Özgür Işık Damar
Back to homeSelected work

Data & Artificial Intelligence Projects

Hybrid search, vector pipelines and agentic systems built to do real work; the open-source ones come with receipts you can check.

All projects

AgentTwin — Data & Artificial Intelligence
Agent assurance2026

Pre-release assurance for tool-using AI agents. AgentTwin runs a baseline and a candidate version against stateful tool twins with injected faults, checks what each run claimed against the twin's final state and names the first step where they diverge. In the bundled demo — a scripted refund agent — a prompt edit that deletes four safety rules to refund faster (the policy check, the idempotency key and two order checks) surfaces as two new critical failures.

GoPythonFastAPINext.js 16+6
Screens (6)Case study
ContextLens — Data & Artificial Intelligence
Topic intelligence2026

A local topic analyser for English conversations. Each message gets one of 8 general topics and a primary subtopic out of 28, plus likely sibling subtopics, with calibrated confidence — or “uncertain” when it sits far from every known topic or confidence is low, and “not English” for other languages. Confident turns build a decaying theme, such as “science books about biology”, that drives a key-less web search.

Python 3.11PyTorchSentence TransformersMiniLM-L6 (fine-tuned)+6
Screens (5)Case study
RealPath — Data & Artificial Intelligence
Relational ML2026

Local-first relational prediction in Python, on DuckDB by default with PostgreSQL and MySQL connectors. RealPath infers keys and time columns across a database's tables, turns a question in English, Turkish or PQL into a predictive task and builds features only from data at or before each prediction date. A score's cross-table features trace back to their join paths: tables, aggregation, value and contribution.

PythonDuckDBPostgreSQLMySQL+6
Screens (5)Case study
Membrane AI — Data & Artificial Intelligence
AI code governance2026

Guardrails for AI-generated code. Each diff runs through a Go pipeline that masks the secrets it detects, applies deterministic checks and can post the verdict as a GitHub commit status and PR comment; the same detectors power an offline CLI that writes an HTML technical-debt report. Model tiers are pluggable and default to a heuristic stand-in, with Claude and Gemini consensus off until enabled.

GoPythonFastAPIgRPC+6
Screens (5)Case study
Hybrid Search Engine — Data & Artificial Intelligence
Data & Artificial Intelligence

Sub-second semantic search across a multilingual product catalogue. Keyword (BM25), dense vector and image embeddings are blended into a single hybrid ranking — customers find products they didn't even know to look for.

QdrantPythonFastAPIHybrid Search+5
Screens (4)
Cross-Market Anomaly Detection & Risk Intelligence — Data & Artificial Intelligence
Data & Artificial Intelligence

Large-scale anomaly detection framework analyzing multi-market transactional and behavioral datasets to identify fraud patterns, pricing manipulation, and irregular activity clusters. Graph-based relationship models and real-time scoring pipelines surface hidden network patterns across sellers, products, and user interactions. Explainable AI for transparent risk evaluation; drift detection keeps accuracy in production.

PythonGraph ModelsReal-time ScoringExplainable AI+1
Screens (1)
Turkish Parliament Transcript Analysis — Filter six thousand pages of parliamentary transcripts by date, keyword, topic, party, city or MP — and watch the national agenda shift in a single chart.
Data & Artificial Intelligence

A platform that uses natural language processing and machine learning to analyse over 6,000 pages of Turkish Parliament transcripts. Explore parliamentary topics, MPs’ speeches and how party priorities change over time, turning lengthy records into accessible, searchable information for citizens and researchers.

PythonNLPMLTurkish Corpus
Screens (3)
Multi-Tenant Data Mesh Architecture — Data & Artificial Intelligence
Data & Artificial Intelligence

Multi-tenant data mesh enabling domain-based data ownership across business units while maintaining federated governance. Data contracts, schema versioning, and cross-domain interoperability rules keep analytics scalable and resilient. Automated quality monitoring and lineage tracking add observability across mission-critical datasets; domain strategies align with business KPIs for faster experimentation.

Data MeshSchema VersioningData ContractsLineage+1
Screens (1)
Real-time Recommendation System — Data & Artificial Intelligence
Data & Artificial Intelligence

Real-time recommendation system using deep learning and reinforcement learning, lifting click-through and engagement on high-volume e-commerce traffic. End-to-end MLOps pipelines on Kubernetes and Docker handle training, deployment and monitoring. Explainable AI surfaces stakeholder trust alongside the metric wins.

Deep LearningReinforcement LearningKubernetesDocker+2
Screens (4)
Seismic Epicenter Prediction — Data & Artificial Intelligence
Data & Artificial Intelligence

ML system analyzing seismic data from 10 Turkish cities affected by major earthquakes, predicting potential future epicenters with advanced models. Geospatial and temporal feature engineering tuned for sparse, high-stakes signal.

PythonMLGeospatialTime Series
Screens (3)
Healthcare ML Pipeline & Dashboards — Data & Artificial Intelligence
Data & Artificial Intelligence

Healthcare data pipeline: preprocessing, visualization and predictive modelling with SVM, Random Forest and Naive Bayes on complex clinical datasets. Interactive dashboards translate raw analysis into business insights for R&D and executive decision-making.

PythonSVMRandom ForestNaive Bayes+1
Screens (4)
Web & Desktop