# Tasaract — Davut Emre Taşar > Start with a question, then follow the links to a project, its evidence and the original source. ## Emre’s Tesseract Welcome in. From the tone of a Turkish sentence to a quantum measurement, these are the questions I keep coming back to. ### [A program’s next move](https://tasaract.com/?scene=project%3Ameaning-native-computing#journey) In Meaning-Native Computing, I ask whether knowing a program’s rules helps a predictor anticipate its next instruction. An independent check decides whether that step can run. [Original source](https://tasaract.com/#meaning-native-computing) ### [The tone of a Turkish review](https://tasaract.com/?scene=repository%3Aturkish-binary-reviews#journey) · 2021 I trained a Turkish BERT model to tell positive reviews from negative ones. In 2022 I released another that could also recognise a neutral review. [Original source](https://github.com/detasar/BERT-TR-Finetuned4-Pos-Neg-Reviews) ### [Listening to Turkish](https://tasaract.com/?scene=work%3Aturkish-asr#journey) · 2024 We taught a speech model trained on 53 languages to transcribe Turkish using recordings from Common Voice. On recordings held back for testing, its word error rate was 23%. [Original source](https://dergipark.org.tr/en/pub/acin/article/1338604) ### [How wide should a forecast be?](https://tasaract.com/?scene=repository%3Aregression-atlas#journey) · 2026 A wide interval can make a forecast look reliable. Across 67 datasets, I compared how often each method caught the actual value and how wide its intervals had to be. [Original source](https://github.com/detasar/regression-conformal-prediction-research-atlas) ### [Who waits for a second look?](https://tasaract.com/?scene=work%3Acoverage-deferral#journey) · 2025 In my simulation on the Adult dataset, calibrating each group separately brought their coverage rates closer. Yet it widened the gap in how often their cases went to human review. [Original source](https://www.preprints.org/manuscript/202512.2631) ### [Two qubits, four outcomes](https://tasaract.com/?scene=work%3Aquantum-conformal#journey) · 2025 Two qubits can give four different answers. I used earlier examples to estimate the uncertainty in all four probabilities together, then tested those estimates on fresh data. [Original source](https://journals.gen.tr/index.php/jqtair/article/view/2702) ### [Customer groups in noisy data](https://tasaract.com/?scene=patent%3A2026-004441#journey) · 2026 How stable are the groups we find in financial data? In application TR 2026/004441, we propose using quantum clustering to form the groups and conformal prediction to quantify the uncertainty. [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) ### [A gap a finger wide](https://tasaract.com/?scene=patent%3A2023-011901#journey) · 2023 Our Turkish filing describes a jointed micro-robot that could carry a camera and a tether into rubble. The same invention appeared in an international patent publication in 2025. [Original source](https://patents.google.com/patent/TR2023011901A2/en) ### [After an agent says “done”](https://tasaract.com/?scene=work%3Aagent-chronicle#journey) · 2026 I built Agent Chronicle so I could retrace a coding session. From a decision I can follow the changes to the files and open the checks behind the result. [Original source](https://github.com/detasar/agent-chronicle-community) ### [Can one model mend another?](https://tasaract.com/?scene=work%3Aweight-editing#journey) · 2026 When I let language models edit a damaged classifier, one run raised balanced accuracy slightly while overall accuracy fell. On intact SmolLM2, the searches kept the original weights. [Original source](https://github.com/detasar/llm-weight-editing-study) ## Where to start - [Can a computer predict which instructions will run next if it knows the rules of the program?](https://tasaract.com/#meaning-native-computing) — Meaning-Native Computing - [Whose case gets a second look?](https://tasaract.com/work/coverage-deferral/) — The coverage–deferral trade-off (Preprint · Preprints.org) - [How much can a finite number of measurements tell us?](https://tasaract.com/work/quantum-conformal/) — Quantum-enhanced conformal methods (Published journal article) - [A coding agent says it is finished. What changed?](https://tasaract.com/work/agent-chronicle/) — Agent Chronicle (Public source project) ## In this edition - [Text guide](https://tasaract.com/guide/) - [Knowledge graph](https://tasaract.com/graph/) - [Profile](https://tasaract.com/profile/) ## About me I’m an engineer and applied scientist on BBVA’s AI Innovation team, part of AI Transformation. I’m doing a PhD at the University of Navarra on fairness in AI. I’m from İzmir, on Turkey’s Aegean coast. I used to be a civil engineer. Then I took an arrow to the knee and traded my hard hat for a lab coat. I’m an engineer and applied scientist on BBVA’s AI Innovation team. At the University of Navarra, I’m doing a PhD on fairness in AI. My interests range from what makes a good team to abstract maths, quantum physics and philosophy. I especially enjoy finding a link between two ideas that seem to have nothing in common. Madrid is home now. I’m exploring the city, learning Spanish and getting to know the cultures of Spain and Latin America. At home I paint, play chess and make up little stories for my daughter. [My favourite espresso machine.](https://tasaract.com/#espresso) [A snapshot from my PhD presentation at the V Jornadas Científicas de DATAI, University of Navarra.](https://tasaract.com/#scholar) I received BBVA’s We Think Big award in 2024 for my work on patents and my academic contributions. · [See the award photo](https://tasaract.com/#we-think-big) - [GitHub](https://github.com/detasar) - [Hugging Face](https://huggingface.co/emre) - [Scholar](https://scholar.google.com/citations?user=IZZM_JAAAAAJ) - [LinkedIn](https://www.linkedin.com/in/detasar/) - [X](https://x.com/mehanixs) - [ORCID](https://orcid.org/0000-0002-7788-0478) - [University of Navarra](https://www.unav.edu/web/instituto-de-ciencia-de-los-datos-e-inteligencia-artificial/equipo#:~:text=Enhancing%20Fairness%20in%20AI%20Systems) ## [Four questions from my notebook](https://tasaract.com/#idea-notebook-title) ### Ontologies and data structures What changes in a graph when a train runs late? A station is a place. A departure is an event. If I give each its own node type and each time a source, can I trace exactly which connections a single delay affects? [RDF 1.1 · W3C](https://www.w3.org/TR/rdf11-concepts/) ### [Meaning-Native Computing](https://tasaract.com/#meaning-native-computing) Can a computer predict which instructions will run next if it knows the rules of the program? Meaning-Native Computing is my independent research project. I want to compare two predictors given the same program and comparable computing resources. One also gets an explicit description of the rules the program must follow. Does that help it predict the next steps? Before a predicted change can take effect, an independent check must confirm that it follows those rules. ### Category theory What survives when I compose two transformations? One process turns a timetable into a graph. Another finds a route through it. When I compose them, how do I preserve which station each time belongs to? [Seven Sketches in Compositionality](https://arxiv.org/abs/1803.05316) ### Quantum measurement What if I ask the same qubit a different question? I can prepare a qubit so it always gives the same answer when measured along one axis. Along a perpendicular axis, either of two results is equally likely. What fascinates me is how changing the measurement changes the answer while the preparation stays the same. [Quantum Information · IBM](https://quantum.cloud.ibm.com/learning/en/courses/basics-of-quantum-information/single-systems/quantum-information) ## Research atlas ### [Uncertainty](https://tasaract.com/#topic-uncertainty) Can intelligence know its limits? Conformal prediction, calibrated uncertainty, and the boundary between a prediction and a decision. - [The coverage–deferral trade-off](https://tasaract.com/work/coverage-deferral/) - [Quantum-enhanced conformal methods](https://tasaract.com/work/quantum-conformal/) - [Regression Conformal Prediction Research Atlas](https://tasaract.com/lab/regression-atlas/) ### [Fairness](https://tasaract.com/#topic-fairness) Who carries the uncertainty? How uncertainty estimates and referral to human review affect different people in decision systems. - [The coverage–deferral trade-off](https://tasaract.com/work/coverage-deferral/) - [Fairness Metrics](https://tasaract.com/lab/fairness-metrics/) ### [Agent systems](https://tasaract.com/#topic-agents) How can you check what an agent actually did? Tools that connect automated actions with evidence, understandable progress, and human oversight. - [Agent Chronicle](https://tasaract.com/work/agent-chronicle/) - [Conversational systems that act](https://tasaract.com/work/chat-assistant/) - [BIRD · Text-to-SQL experiments](https://tasaract.com/lab/bird-sql/) ### [Quantum](https://tasaract.com/#topic-quantum) What can a measurement tell us? Statistical calibration, anomaly detection, and adversarial evaluation around quantum measurements. - [Quantum-enhanced conformal methods](https://tasaract.com/work/quantum-conformal/) - [TARA · Test by Adaptive Ranks](https://tasaract.com/work/tara/) - [Quantum Time Experiment](https://tasaract.com/lab/quantum-clocks/) ### [Invention](https://tasaract.com/#topic-invention) How small would a rescue robot need to be? Patent applications on rescue robotics, conversational interfaces, personal knowledge graphs and quantum methods. - [Conversational systems that act](https://tasaract.com/work/chat-assistant/) - [Micro-robots for search and rescue](https://tasaract.com/work/micro-robot-rescue/) ### [Language](https://tasaract.com/#topic-language) How do machines understand us? Turkish speech recognition, language-model adaptation, and contributions to open multilingual research. - [Conversational systems that act](https://tasaract.com/work/chat-assistant/) - [AI-assisted point of sale](https://tasaract.com/work/ai-pos/) - [Turkish Sentiment Analysis · TSAM](https://tasaract.com/lab/turkish-sentiment/) ## Research note — [The coverage–deferral trade-off](https://tasaract.com/work/coverage-deferral/) Whose case gets a second look? Table 4 · Percentage points. 100 random seeds. Target coverage: 90%. | Dataset | Coverage gap · Pooled calibration | Coverage gap · Groupwise | Referral gap · Pooled calibration | Referral gap · Groupwise | | --- | ---: | ---: | ---: | ---: | | Adult | 5.0 | 1.2 | 5.1 | 12.9 | | ACS Income | 0.8 | 0.8 | 3.4 | 5.4 | | Bank Marketing | 2.0 | 1.1 | 0.0 | 3.0 | | COMPAS | 1.8 | 3.3 | 5.7 | 10.8 | | German Credit | 3.5 | 5.7 | 6.1 | 24.7 | | Taiwan Credit | 1.1 | 1.3 | 3.4 | 7.7 | [Original source](https://www.preprints.org/manuscript/202512.2631#:~:text=Table%204.%20Coverage-deferral%20trade-off%20across%20datasets) ## Research & publications ### [The coverage–deferral trade-off](https://tasaract.com/work/coverage-deferral/) Original title · The Coverage-Deferral Trade-Off: Fairness Implications of Conformal Prediction in Human-in-the-Loop Decision Systems 2025 · Preprint · Preprints.org · Sole author · Davut Emre Taşar · Preprints.org Whose case gets a second look? A study of how conformal prediction changes both uncertainty coverage and who is referred for human review. A prediction set can contain both possible answers. In this study, those cases go to a human reviewer. I compared a single calibration threshold with thresholds fitted separately for each group. The choice of threshold changes how often the right answer is in the set and how often a person has to wait for review. On the Adult income dataset, groupwise calibration narrowed the gap in coverage from 5.0 to 1.2 percentage points. The gap in referral to a human grew from 5.1 to 12.9 points. Table 4 reports these results over 100 random seeds, with 90% target coverage and simulated review. [Fairness](https://tasaract.com/#topic-fairness) · [Uncertainty](https://tasaract.com/#topic-uncertainty) [Original source](https://www.preprints.org/manuscript/202512.2631) ### [Agent Chronicle](https://tasaract.com/work/agent-chronicle/) 2026 · Public source project · Project maintainer · Agent Chronicle Community · GitHub A coding agent says it is finished. What changed? A local control surface for coding agents that connects progress, decisions, changes, and verification to recorded evidence. Agent Chronicle turns a coding session into a timeline you can inspect. Follow a Codex or Claude task from its original goal through decisions, file changes and checks, then open the recorded activity behind a claim. The public repository includes a local demo, setup guides and an Apache 2.0 licence for the project’s own code. Its interface is available in English, Turkish and Spanish. [Agent systems](https://tasaract.com/#topic-agents) [Original source](https://github.com/detasar/agent-chronicle-community) ### [Quantum-enhanced conformal methods](https://tasaract.com/work/quantum-conformal/) Original title · Quantum-Enhanced Conformal Methods for Multi-Output Uncertainty: A Holistic Exploration and Experimental Analysis 2025 · Published journal article · Sole author · Davut Emre Taşar · Journal of Quantum Technologies and Informatics Research How much can a finite number of measurements tell us? An investigation of conformal uncertainty estimation for multi-output distributions arising from quantum measurements. A quantum experiment produces counts for several possible outcomes. I studied how to estimate their probabilities together and how wide a set of plausible answers should be. The method calibrates predictions against earlier examples and checks the result on held-out data. The 2025 journal article evaluates both coverage and set size on simulated data and measurements in several bases. The accompanying notebooks show how to rerun the experiments. [Uncertainty](https://tasaract.com/#topic-uncertainty) · [Quantum](https://tasaract.com/#topic-quantum) [Original source](https://journals.gen.tr/index.php/jqtair/article/view/2702) ### [TARA · Test by Adaptive Ranks](https://tasaract.com/work/tara/) Original title · TARA Test-by-Adaptive-Ranks for Quantum Anomaly Detection with Conformal Prediction Guarantees 2025 · arXiv preprint · First author · Davut Emre Taşar, Ceren Öcal Taşar · arXiv When does a stream of measurements start to look unusual? A framework combining conformal prediction and sequential testing to study anomalies in quantum correlations. Ceren Öcal Taşar and I studied how to flag changes in quantum correlations as observations arrive. One version compares a batch with a reference distribution. A second accumulates evidence over successive measurements. The preprint reports experiments on IBM Torino and IonQ Forte Enterprise. It also examines how using the same distribution for calibration and evaluation can change the apparent performance. [Quantum](https://tasaract.com/#topic-quantum) · [Uncertainty](https://tasaract.com/#topic-uncertainty) [Original source](https://arxiv.org/abs/2512.04016) ### [Conversational systems that act](https://tasaract.com/work/chat-assistant/) Original title · A chat assistant system 2025 · Published PCT application · Co-inventor · WIPO · PCT When should a conversation turn into an action? An invention proposal for an AI-assisted conversational system combining information retrieval with command-driven actions. The proposed assistant can search for information when a user asks a question and carry out an operation when given a command. The application puts both kinds of interaction in a single conversation. The Turkish filing is TR 2023/011870. The international application was published on 3 April 2025. [Invention](https://tasaract.com/#topic-invention) · [Agent systems](https://tasaract.com/#topic-agents) · [Language](https://tasaract.com/#topic-language) [Original source](https://patents.google.com/patent/WO2025071524A1/en) ### [Micro-robots for search and rescue](https://tasaract.com/work/micro-robot-rescue/) Original title · A system for providing search and rescue operation via a micro-robot 2025 · Published PCT application · Co-inventor · WIPO · PCT What could reach someone through a gap only a finger wide? A proposed micro-robot system for locating victims and supporting rescue operations in confined environments. Our application proposes a finger-sized, jointed robot for narrow spaces in rubble. A camera and sensors would help locate a trapped person. A tether would carry a communications and support line through the same opening. Türkiye Garanti Bankası filed TR 2023/011901 on 25 September 2023. The international publication WO2025071525A1 followed on 3 April 2025. [Invention](https://tasaract.com/#topic-invention) · [Agent systems](https://tasaract.com/#topic-agents) [Original source](https://patents.google.com/patent/WO2025071525A1/en) ### [AI-assisted point of sale](https://tasaract.com/work/ai-pos/) Original title · An artificial intelligence-aided point of sale device management system 2024 · Published PCT application · Co-inventor · WIPO · PCT Could a shopkeeper operate a payment terminal by speaking? A proposed POS system connecting personalized recommendations, fraud detection, and natural-language device operation. This application describes a merchant operating a POS terminal remotely with spoken or written instructions. It also proposes generating QR-coded output and analysing transactions for potential fraud. The international application WO2024220062A1 was published on 24 October 2024. The related Turkish filing is TR 2023/004244. [Invention](https://tasaract.com/#topic-invention) · [Language](https://tasaract.com/#topic-language) [Original source](https://patents.google.com/patent/WO2024220062A1/en) ### [Gesture-based virtual interaction](https://tasaract.com/work/virtual-input/) Original title · A system for using a virtual keyboard and a mouse 2025 · Published PCT application · Co-inventor · WIPO · PCT Where is the keyboard when the screen is virtual? An invention proposal using hand-motion tracking for interaction with virtual and augmented reality environments. The proposal uses camera observations of hand movement as input in virtual and augmented reality. The system interprets those movements as keyboard or mouse actions. The Turkish record TR 2023/011601 and the PCT publication WO2025063923A1 describe the same invention. The PCT application was published on 27 March 2025. [Invention](https://tasaract.com/#topic-invention) [Original source](https://patents.google.com/patent/WO2025063923A1/en) ### [Personalized workplace suggestions](https://tasaract.com/work/personalized-suggestions/) Original title · A system for increasing employee productivity by providing personalized suggestions 2025 · Published PCT application · Co-inventor · WIPO · PCT How could a workplace assistant tailor a suggestion? A proposed system generating personalized workplace recommendations from daily mood and health signals. The application describes a system that generates suggestions using signals about an employee’s day, mood and health. The document sets out a proposed design for workplace recommendations. The related Turkish filing is TR 2023/011830. The international publication WO2025063924A1 lists Türkiye Garanti Bankası as applicant. [Invention](https://tasaract.com/#topic-invention) [Original source](https://patents.google.com/patent/WO2025063924A1/en) ### [Adversarial limits of quantum certification](https://tasaract.com/work/adversarial-quantum/) Original title · Adversarial Limits of Quantum Certification: When Eve Defeats Detection 2025 · arXiv preprint · Sole author · Davut Emre Taşar · arXiv Does a good test result hold up under different conditions? An exploration of adaptive adversaries and evaluation limits in statistical quantum certification. I studied how the choice of calibration and evaluation data affects reported performance in tests of quantum correlations. The experiment asks when a statistical score is a reliable piece of evidence. The December 2025 preprint describes the simulation, the comparison with IBM Quantum measurements and the assumptions behind the evaluation. [Quantum](https://tasaract.com/#topic-quantum) · [Uncertainty](https://tasaract.com/#topic-uncertainty) [Original source](https://arxiv.org/abs/2512.04391) ### [Turkish speech recognition](https://tasaract.com/work/turkish-asr/) Original title · Transformer-Based Turkish Automatic Speech Recognition 2024 · Published journal article · First author · Davut Emre Taşar, Kutan Koruyan, Cihan Çılgın · Acta Infologica What carries over when a speech model learns Turkish? An evaluation of self-supervised speech representations and transfer learning for Turkish speech recognition. With Kutan Koruyan and Cihan Çılgın, I adapted a Wav2Vec 2.0 model pretrained on speech in 53 languages. We used recordings and transcripts from Mozilla Common Voice Turkish to teach the model to turn Turkish speech into text. The paper reports a 23% word error rate on the Common Voice Turkish test split. This measure counts substitutions, deletions and insertions relative to the length of the reference transcript. [Language](https://tasaract.com/#topic-language) [Original source](https://dergipark.org.tr/en/pub/acin/article/1338604) ### [Qwen-0.5B-GRPO](https://tasaract.com/work/qwen-grpo/) Original title · Qwen-0.5B-GRPO: A Fine-Tuned Math Reasoner 2025 · Public model adaptation · Model developer · Davut Emre Taşar · Hugging Face Models Can a small model learn to lay out a maths answer? A GRPO fine-tune of Qwen2.5-0.5B-Instruct for structured mathematical reasoning using GSM8K. I adapted Qwen2.5-0.5B-Instruct on GSM8K word problems. The fine-tuning rewards a correct numerical answer and a response that follows the requested structure. These properties are scored separately. The model card documents a single training epoch, the upstream Qwen model and example inference code. The public weights let other researchers try the adaptation themselves. [Language](https://tasaract.com/#topic-language) [Original source](https://huggingface.co/emre/Qwen-0.5B-GRPO) ### [TARA · Reasoning in Turkish](https://tasaract.com/work/turkish-reasoning-benchmark/) Original title · TARA: Turkish Advanced Reasoning Assessment Dataset 2025 · Public research dataset · Version 1 · Co-creator · Davut Emre Taşar, Ceren Öcal Taşar · Hugging Face Datasets How do you ask a language model to reason in Turkish? A set of Turkish-language tasks for examining how language models reason across domains. Ceren Öcal Taşar and I assembled tasks across ten categories, including logic, planning, Python and SQL. A reader can inspect the prompts and expected responses before looking at a model’s score. Version 1 contains 100 synthetic examples. The dataset card describes their generation and notes that a single generator and a small sample limit what the scores say about everyday Turkish. [Language](https://tasaract.com/#topic-language) [Original source](https://huggingface.co/datasets/emre/TARA_Turkish_LLM_Benchmark) ### [One language model for several chatbots](https://tasaract.com/work/multilingual-chatbots/) Original title · Performance Trade-Off for Bert Based Multi-Domain Multilingual Chatbot Architectures 2021 · Peer-reviewed journal article · First and corresponding author · Davut Emre Taşar, Şükrü Ozan, Seçilay Kutal, Oğuzhan Ölmez, Semih Gülüm, Fatih Akca, Ceren Belhan · Journal of Artificial Intelligence and Data Science Do thirteen chatbot tasks need thirteen models? We tested whether separate chatbots could share one multilingual BERT model while keeping their own sets of possible answers. A visitor asking how to get a virtual card and a visitor asking about an unrelated subject need different answers. Our classifier shares the language model across chatbots and uses the active chatbot to narrow the possible responses. The evaluation included banking questions covering 77 intents. Across the study we tested 13 classification tasks and 333 classes in Turkish, English and German. In the five-model comparison, the separate models used 10 GB of memory in total. The shared model used 2 GB, with a small loss of accuracy. [Language](https://tasaract.com/#topic-language) [Original source](https://dergipark.org.tr/en/pub/jaida/article/1060221) ### [Finding mismatched labels in ad copy](https://tasaract.com/work/ad-text-classification/) Original title · BERT Modeli’nin Sınıflandırma Doğruluğunun Sıfır-Atış Öğrenmesi ile Artırılması 2021 · Peer-reviewed journal article · Co-author · Şükrü Ozan, Umut Özdil, Davut Emre Taşar, Büşra Arslan, Gökçe Polat · Türkiye Bilişim Vakfı Bilgisayar Bilimleri ve Mühendisliği Dergisi What if the label on an ad is the mistake? We examined how mismatches between an advertisement and its industry label affect a BERT text classifier. The classifier struggled in particular industries. We used a zero-shot model to check whether each ad matched its assigned sector, cleaned the training data, then trained BERT again. The original corpus contained about 101,000 labelled ads from 44 industries. The article reports improved classification performance after this preprocessing step. [Language](https://tasaract.com/#topic-language) [Original source](https://dergipark.org.tr/tr/pub/tbbmd/article/1004781) ### [Editing neural networks with language models](https://tasaract.com/work/weight-editing/) Original title · Editing Neural Networks with Language Models 2026 · Public experiment, report and code · Sole author · Davut Emre Taşar · GitHub Can one model edit the weights of another? I asked language models to propose small weight changes for two different tasks. One target was a damaged classifier. The other was an intact language model. In the UCI Adult income classifier I permanently disabled some weights. Sol and Astra could propose edits to the surviving weights, then a joint search could use their development histories. In a separate experiment I tested edits on intact SmolLM2-1.7B using ARC-Easy science questions and PIQA physical-reasoning questions. In both experiments, keeping the original weights was an eligible option. On 16,281 held-out Adult examples, Sol’s edit raised balanced accuracy from 78.59% to 79.02%. Overall accuracy fell from 82.86% to 82.63%, and probability estimates worsened. On SmolLM2 every completed search selected the original weights. These are exploratory results from one run. [Agent systems](https://tasaract.com/#topic-agents) · [Language](https://tasaract.com/#topic-language) [Original source](https://github.com/detasar/llm-weight-editing-study) ### [BLOOM](https://tasaract.com/work/bloom/) Original title · BLOOM: A 176B-Parameter Open-Access Multilingual Language Model 2022 · Collaborative research paper · Co-author · BigScience Workshop · BigScience Workshop · arXiv How was a multilingual model built in the open? A contribution to the BigScience collaboration on open multilingual language-model research. BigScience brought together a large international research collaboration to document the training and release of BLOOM, a multilingual language model. I am one of the co-authors of the project paper. The BigScience paper gives the full author list and describes the architecture, data and training process. [Language](https://tasaract.com/#topic-language) [Original source](https://arxiv.org/abs/2211.05100) ## Patents & utility models Public registry and publication records name Emre Taşar as an inventor on 28 patent applications and 2 utility model applications in Türkiye. Filing numbers, titles and dates come from the TÜRKPATENT extract of 2026-08-31, with protection types from the inventor’s list. Current legal status is checked separately in the register. Some inventions also have an international publication. National filings may overlap with PCT publications. The total counts applications filed in Türkiye. Portfolio-wide legal status is not provided by the dated extract. Individual grant decisions cite the national register. Applicant · TÜRKİYE GARANTİ BANKASI ANONİM ŞİRKETİ Register snapshot · 2026-08-31 · [Official TÜRKPATENT register](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) ### [2026/007909 · Noise reduction system based on the user's voice](https://tasaract.com/patents/2026-007909/) Patent application · Filed 2026-05-18 The system isolates a speaker’s voice from background sounds and transcribes it. It can edit the wording to suit the speaker’s preferences or the setting, then regenerate the speech with the speaker’s vocal characteristics. Published · 2026-07-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/007909 ### [2026/007907 · Logical inference-based artificial intelligence system](https://tasaract.com/patents/2026-007907/) Patent application · Filed 2026-05-18 The application describes turning transaction histories, user records and network connections into a logical knowledge model. The system can then draw inferences from those records in fields such as finance, health and logistics. Published · 2026-06-22 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/007907 ### [2026/006294 · A personalized waiting experience system](https://tasaract.com/patents/2026-006294/) Patent application · Filed 2026-04-24 The proposed system assesses a person’s emotional state while they wait and selects content for them based on that assessment. Published · 2026-05-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/006294 ### [2026/006289 · A holistic retirement planning and life assistance system supported by a personalized dynamic ontology and knowledge graph](https://tasaract.com/patents/2026-006289/) Patent application · Filed 2026-04-24 The application proposes a personal ontology and a knowledge graph that can evolve over time. It brings those representations together to support retirement planning and offer proactive assistance in everyday life. Published · 2026-06-22 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/006289 ### [2026/004444 · An AI-supported and privacy-focused environment sensing system](https://tasaract.com/patents/2026-004444/) Patent application · Filed 2026-03-26 The application addresses spaces where cameras are restricted or unsuitable. It proposes detecting whether people are present or active while preserving privacy, so that energy use can be adjusted to the situation. Published · 2026-05-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/004444 ### [2026/004441 · A noise-resistant dynamic financial customer segmentation system using quantum entanglement and conformal prediction](https://tasaract.com/patents/2026-004441/) Patent application · Filed 2026-03-26 The application proposes dynamic financial customer segments that can tolerate noisy data. It brings together quantum entanglement to represent complex patterns, variational quantum clustering to form groups, and conformal prediction to provide statistical assurance. Published · 2026-04-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/004441 ### [2026/002909 · A prediction system](https://tasaract.com/patents/2026-002909/) Patent application · Filed 2026-02-26 The system is intended to give a numerical measure of confidence in an AI prediction. The abstract gives examples including a borrower defaulting, a transaction producing a profit or loss, and an operation finishing late. Published · 2026-03-23 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/002909 ### [2026/001992 · A system providing customer identification, time measurement and experience optimization in banks using computer imagery and artificial intelligence](https://tasaract.com/patents/2026-001992/) Patent application · Filed 2026-02-10 The proposed system uses existing computer-vision and AI algorithms to identify customers in a bank branch, measure visit durations and estimate waiting times. It also describes monitoring departures and adjusting the queue. Published · 2026-03-23 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026/001992 ### [2025/019989 · A gamification system](https://tasaract.com/patents/2025-019989/) Patent application · Filed 2025-12-12 The system analyses banking activity and proposes generating financial content and game scenarios tailored to the user. The banking application can collect feedback on those scenarios. Published · 2026-01-21 [Original source](https://patents.google.com/patent/TR2025019989A2/en) · TR2025019989A2 ### [2025/011133 · A system enabling cognitive enrichment of captured image content with federated learning support](https://tasaract.com/patents/2025-011133/) Patent application · Filed 2025-08-08 The filing describes personalising audiovisual memories captured on a user’s device, with federated learning as part of the proposed design. Published · 2025-09-22 [Original source](https://patents.google.com/patent/TR2025011133A2/en) · TR2025011133A2 ### [2025/008786 · A probabilistic and dynamically extensible graph-based variant generation system for multimodal content](https://tasaract.com/patents/2025-008786/) Patent application · Filed 2025-06-30 The system extracts features from sound, video or images, then builds a graph of possible variations. Branches can represent alternative pasts, futures or parallel versions of the original content. Published · 2025-08-21 [Original source](https://patents.google.com/patent/TR2025008786A2/en) · TR2025008786A2 ### [2025/008273 · A personal transaction booth system](https://tasaract.com/patents/2025-008273/) Patent application · Filed 2025-06-20 A modular service booth gives a person a private place to handle sensitive interactions with several service providers. Published · 2025-09-22 [Original source](https://patents.google.com/patent/TR2025008273A2/en) · TR2025008273A2 ### [2025/008055 · A driving safety system](https://tasaract.com/patents/2025-008055/) Patent application · Filed 2025-06-18 The proposed system uses sensors and AI to detect and assess a vehicle collision as it occurs. It can notify the relevant services and suggest transport or accommodation to the people involved. Published · 2025-08-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2025/008055 ### [2025/008032 · A collective decision-making system using human–AI hybrid swarm intelligence](https://tasaract.com/patents/2025-008032/) Patent application · Filed 2025-06-17 The application brings human and AI participants into a collective decision process. It proposes several levels of reliability assessment within a system designed to adapt and improve over time. Published · 2025-07-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2025/008032 ### [2025/008028 · A coordination system in a multi-device ecosystem](https://tasaract.com/patents/2025-008028/) Patent application · Filed 2025-06-17 The proposed system uses AI to coordinate devices and applications within an ecosystem. Its stated aim is a user experience that can evolve as the person interacts with that ecosystem. Published · 2025-07-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2025/008028 ### [2025/007997 · A holistic real estate valuation analysis and strategic planning system](https://tasaract.com/patents/2025-007997/) Patent application · Filed 2025-06-17 This application combines data from multiple sources for real-estate valuation. It proposes using an evolving ontology, network analysis and explainable generative AI to connect valuation with analysis and strategic planning. Published · 2025-07-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2025/007997 ### [2025/007995 · A development environment system with an agent architecture and related data format](https://tasaract.com/patents/2025-007995/) Patent application · Filed 2025-06-17 The application proposes an agent architecture and an accompanying data format for building AI assistants. The agents are designed to move between platforms and adapt over time while retaining a personalized experience. Published · 2025-08-21 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2025/007995 ### [2024/003815 · A system improving customer communication in the financial sector](https://tasaract.com/patents/2024-003815/) Patent application · Filed 2024-03-27 The proposal connects mobile banking, augmented reality and virtual branches so customers can discuss investments or other financial topics with each other and interact with AI models. Published · 2024-04-22 [PCT · WO2025207066A1](https://patents.google.com/patent/WO2025207066A1/en) [Original source](https://patents.google.com/patent/TR2024003815A2/en) · TR2024003815A2 ### [2024/003813 · A system enabling the creation of investment channels](https://tasaract.com/patents/2024-003813/) Patent application · Filed 2024-03-27 The application proposes investment channels curated by experts or other public figures. It describes AI avatars based on particular profiles delivering financial guidance to the channel’s users. Published · 2024-04-22 [Original source](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2024/003813 ### [2024/003592 · A system enabling people to help one another with certain banking transactions through mutual consent](https://tasaract.com/patents/2024-003592/) Patent application · Filed 2024-03-25 A customer who wants help with a banking transaction can complete it with someone they trust. The proposal uses mutual approval to let the two people take part together. Published · 2024-05-21 [PCT · WO2025207061A1](https://patents.google.com/patent/WO2025207061A1/en) [Original source](https://patents.google.com/patent/TR2024003592A2/en) · TR2024003592A2 ### [2023/011901 · A system providing search and rescue operations through a micro-robot](https://tasaract.com/patents/2023-011901/) Patent application · Filed 2023-09-25 Micro-robots would enter confined spaces in a collapsed building or mine, gather information about the surroundings and help locate trapped people. The system also proposes a line for delivering support. Published · 2023-12-21 [Final Turkish patent-grant decision announced](https://www.turkpatent.gov.tr/arastirma-yap?form=patent) · 2026-02-23 [PCT · WO2025071525A1](https://patents.google.com/patent/WO2025071525A1/en) [Original source](https://patents.google.com/patent/TR2023011901A2/en) · TR2023011901A2 ### [2023/011870 · A chat assistant system](https://tasaract.com/patents/2023-011870/) Patent application · Filed 2023-09-25 The proposed assistant can look up information online when a user asks a question and perform an operation when the user gives a command. Both interactions take place in the same conversation. Published · 2023-12-21 [PCT · WO2025071524A1](https://patents.google.com/patent/WO2025071524A1/en) [Original source](https://patents.google.com/patent/TR2023011870A2/en) · TR2023011870A2 ### [2023/011830 · A system increasing employee productivity through personalized suggestions](https://tasaract.com/patents/2023-011830/) Patent application · Filed 2023-09-22 The proposed workplace assistant uses daily information about an employee’s mood and health to generate individual suggestions for the working day. Published · 2024-08-21 [PCT · WO2025063924A1](https://patents.google.com/patent/WO2025063924A1/en) [Original source](https://patents.google.com/patent/TR2023011830A2/en) · TR2023011830A2 ### [2023/011601 · A system enabling use of a virtual keyboard and mouse](https://tasaract.com/patents/2023-011601/) Patent application · Filed 2023-09-18 The system tracks hand movements and interprets them as keyboard or mouse input in virtual and augmented reality. Published · 2023-12-21 [PCT · WO2025063923A1](https://patents.google.com/patent/WO2025063923A1/en) [Original source](https://patents.google.com/patent/TR2023011601A2/en) · TR2023011601A2 ### [2023/011600 · A system increasing security in ATM transactions](https://tasaract.com/patents/2023-011600/) Utility model application · Filed 2023-09-18 The system watches for a card, cash or receipt left at an ATM after a transaction. The banking app can notify a customer who has walked away. Published · 2024-01-22 [PCT · WO2025063922A1](https://patents.google.com/patent/WO2025063922A1/en) [Original source](https://patents.google.com/patent/TR2023011600A2/en) · TR2023011600A2 ### [2023/011599 · A system facilitating ATM use](https://tasaract.com/patents/2023-011599/) Patent application · Filed 2023-09-18 A phone or tablet reads a code on an ATM and displays its controls in augmented reality. The physical machine handles functions such as accepting and dispensing cash and printing receipts. Published · 2024-01-22 [PCT · WO2025063920A1](https://patents.google.com/patent/WO2025063920A1/en) [Original source](https://patents.google.com/patent/TR2023011599A2/en) · TR2023011599A2 ### [2023/004244 · An AI-supported point-of-sale device management system](https://tasaract.com/patents/2023-004244/) Patent application · Filed 2023-04-17 The proposal lets a merchant control a payment terminal remotely using speech or written instructions. It also describes QR-coded output and analysis of transactions for signs of fraud. Published · 2023-09-21 [PCT · WO2024220062A1](https://patents.google.com/patent/WO2024220062A1/en) [Original source](https://patents.google.com/patent/TR2023004244A2/en) · TR2023004244A2 ### [2023/004047 · A secure shopping system](https://tasaract.com/patents/2023-004047/) Patent application · Filed 2023-04-12 Parties to a credit or letter-of-credit transaction can create or join a contract. The system records its stages and supporting documents on a blockchain and uses AI to suggest matches between the parties. Published · 2023-09-21 [PCT · WO2024215289A1](https://patents.google.com/patent/WO2024215289A1/en) [Original source](https://patents.google.com/patent/TR2023004047A2/en) · TR2023004047A2 ### [2022/016679 · A recommendation system](https://tasaract.com/patents/2022-016679/) Utility model application · Filed 2022-11-03 The proposal derives product suggestions from text, images, audio or movement supplied in a conversation or virtual environment. It keeps previous inputs and recommendations so the exchange can continue across interactions. Published · 2022-12-21 [PCT · WO2024096828A1](https://patents.google.com/patent/WO2024096828A1/en) [Original source](https://patents.google.com/patent/WO2024096828A1/en) · WO2024096828A1 ### [2022/016395 · A system managing the voting process](https://tasaract.com/patents/2022-016395/) Patent application · Filed 2022-10-31 The voting system brings several AI methods into a collective decision process. The proposal addresses how participants communicate as a vote unfolds. Published · 2022-12-21 [PCT · WO2024096827A1](https://patents.google.com/patent/WO2024096827A1/en) [Original source](https://patents.google.com/patent/TR2022016395A2/en) · TR2022016395A2 ## Models & code ### [Turkish Sentiment Analysis · TSAM](https://tasaract.com/lab/turkish-sentiment/) Model · AutoTrain classifier I trained this BERT model to read the tone of a Turkish review. It scores the sentence for positive, neutral and negative sentiment. An independent study tested four models on four Turkish datasets. In its binary evaluation, the researchers excluded neutral labels. This model scored 0.97 F1 on Humir and 0.54 on Twt. [A cross-validation study of Turkish sentiment analysis datasets and tools](https://doi.org/10.1007/s10579-025-09869-6) [Total Hub downloads](https://huggingface.co/api/models/emre/turkish-sentiment-analysis?expand=downloadsAllTime) · 12,053 · 2026-11-18 [An earlier review classifier](https://github.com/detasar/BERT-TR-Finetuned4-Pos-Neg-Reviews) I fine-tuned Turkish BERT to tell positive reviews from negative ones. The training settings are on GitHub. [Original source](https://huggingface.co/emre/turkish-sentiment-analysis) ### [Whisper Medium TR](https://tasaract.com/lab/whisper-tr/) Model · Speech-model adaptation Give it a Turkish recording and it returns a transcript. I adapted Whisper Medium on Mozilla Common Voice 11.0 speech to study recognition errors in Turkish. [OpenAI Whisper Medium](https://huggingface.co/openai/whisper-medium) [Original source](https://huggingface.co/emre/whisper-medium-turkish-2) ### [Gemma 2 · Turkish Wikipedia](https://tasaract.com/lab/gemma-wiki-tr/) Model · LoRA adapter author This adapter continues a Turkish passage using Gemma 2 9B trained further on Turkish Wikipedia. The card permits academic and research use and limits commercial use to BBVA Group. [Google Gemma 2 9B](https://huggingface.co/google/gemma-2-9b) [Original source](https://huggingface.co/emre/gemma-2-9b-Turkish-Lora-Continue-Pre-Trained) ### [Java Methods · Doc2Vec](https://tasaract.com/lab/java-method-vectors/) Model · Embedding model developer A Java method goes in and a 200-dimensional vector comes out. Developers can use these vectors to find similar snippets or group methods in a codebase. [Gensim Doc2Vec](https://radimrehurek.com/gensim/models/doc2vec.html) [Original source](https://huggingface.co/emre/java_8m_methods_doc2vec) ### [Gemma 3 27B · Turkish reasoning](https://tasaract.com/lab/gemma-reasoning-tr/) Model · Fine-tune of a quantized model I fine-tuned Unsloth’s 4-bit build of Gemma 3 27B Instruct for answering reasoning questions in Turkish. The card includes preliminary TARA evaluations on tasks such as logic, planning and SQL. [Unsloth 4-bit Gemma 3 27B Instruct](https://huggingface.co/unsloth/gemma-3-27b-it-unsloth-bnb-4bit) [Original source](https://huggingface.co/emre/gemma-3-27b-it-tr-reasoning40k-4bit) ### [arXiv Abstract-to-Title · mT5](https://tasaract.com/lab/abstract-title/) Model · Sequence-to-sequence fine-tune Paste a research abstract to get a candidate paper title. This mT5-small adaptation learned the task from abstract–title pairs in an arXiv dataset. [mT5-small](https://huggingface.co/google/mt5-small) [Original source](https://huggingface.co/emre/arxiv27k-t5-abst-title-gen) ### [Wav2Vec2 XLS-R · Turkish](https://tasaract.com/lab/xlsr-tr/) Model · Speech-model adaptation This Turkish speech recogniser turns recorded audio into text using the multilingual XLS-R encoder. Researchers can inspect its training history and its word error rate on Turkish Common Voice. [Meta Wav2Vec2 XLS-R 300M](https://huggingface.co/facebook/wav2vec2-xls-r-300m) [Original source](https://huggingface.co/emre/wav2vec2-large-xls-r-300m-tr) ### [Qwen 0.5B · GRPO maths](https://tasaract.com/lab/qwen-grpo/) Model · Reinforcement-learning fine-tune A small model for maths word problems, trained on GSM8K to produce a worked response and a numerical answer. The training rewards answer correctness and response format separately. [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) [Original source](https://huggingface.co/emre/Qwen-0.5B-GRPO) ### [Regression Conformal Prediction Research Atlas](https://tasaract.com/lab/regression-atlas/) Repository · Study author Look up a dataset and a method to see how often its prediction intervals covered the outcome and how wide those intervals were. The atlas links results across 67 datasets to experiments and a browsable knowledge graph. [Original source](https://github.com/detasar/regression-conformal-prediction-research-atlas) ### [BIRD · Text-to-SQL experiments](https://tasaract.com/lab/bird-sql/) Repository · Research repository maintainer The pipeline takes a question and a database schema, proposes SQL and checks the result by executing it. Researchers can inspect how ambiguity, table selection and query validation affect the answer on a financial subset of BIRD. [Original source](https://github.com/detasar/bird-text2sql-experiments) ### [WebCLI](https://tasaract.com/lab/webcli/) Repository · Project maintainer WebCLI indexes public documentation and lets developers or agents search it locally from the command line. Results include stable references and can be exported as JSONL, with OpenAI’s developer documentation as the first adapter. [Original source](https://github.com/detasar/webcli) ### [Fairness Metrics](https://tasaract.com/lab/fairness-metrics/) Repository · Library maintainer Give the Python library features, predictions, actual outcomes and a protected group. It returns a pandas table comparing quantities such as false-positive, false-negative and recall rates across groups. [Original source](https://github.com/detasar/fairness_metrics) ### [Quantum Time Experiment](https://tasaract.com/lab/quantum-clocks/) Repository · Research repository maintainer If several fragments of a quantum system preserve the same record, can they tell us what happened earlier? The experiments test which records can be put in order and when an additional reference is needed to determine the direction. [Original source](https://github.com/detasar/quantum-time-experiment) ### [PaperBanana field test](https://tasaract.com/lab/paperbanana-test/) Repository · Independent evaluation author I tested Zhu and colleagues’ illustration pipeline on the methods sections of three of my papers. The repository pairs each untouched output with the original figure and exact input, so readers can judge where the illustrations went wrong. [Original source](https://github.com/detasar/paperbanana-field-test) ### [World Cup 2026 Nowcast Lab](https://tasaract.com/lab/worldcup-nowcast/) Repository · Forecasting project maintainer This research lab turns match probabilities into distributions over possible tournament winners by simulating the 2026 World Cup. Readers can inspect the forecast card, data sources and model choices behind each snapshot. [Original source](https://github.com/detasar/worldcup26-nowcast-lab) ### [Agent Chronicle](https://tasaract.com/lab/agent-chronicle/) Repository · Project maintainer Read a Codex or Claude session as a timeline of decisions, changes and checks. Each recorded event gives the developer a way to inspect the evidence behind a progress report. [Original source](https://github.com/detasar/agent-chronicle-community) ## Source documents - [Profile](https://tasaract.com/profile.json) - [Research & publications](https://tasaract.com/public-works.json) - [Patents & utility models](https://tasaract.com/patent-portfolio.json) - [Models & code](https://tasaract.com/open-lab.json) - [Knowledge graph](https://tasaract.com/ontology.jsonld) - [Choose a path](https://tasaract.com/journey-data.json) - [JSON](https://tasaract.com/agent-index.json) ## API · MCP - [MCP · JSON-RPC · POST](https://tasaract.com/mcp?lang=en) - [Profile](https://tasaract.com/api/profile?lang=en) - [Research & publications](https://tasaract.com/api/works?lang=en) - [Patents & utility models](https://tasaract.com/api/patents?lang=en) - [Models & code](https://tasaract.com/api/lab?lang=en) - [Language](https://tasaract.com/languages.json)