# Umut Dinçer Yananer > Software engineer who builds across backend and data/ML. - Site: https://yananer.dev - Agent skill file: https://yananer.dev/SKILL.md - Résumé (JSON Resume): https://yananer.dev/resume.json - GitHub: https://github.com/umutdinceryananer - LinkedIn: https://www.linkedin.com/in/umut-yananer/ ## Public projects - [Government-Citizen-Services-Voice-Agent](https://github.com/umutdinceryananer/Government-Citizen-Services-Voice-Agent): A multilingual (TR/EN) AI voice agent for government services: a LangGraph agent over a bilingual RAG knowledge base, with ElevenLabs speech, a FastAPI + Twilio phone layer, and a Streamlit analytics dashboard. - [FX-Risk-Engine](https://github.com/umutdinceryananer/FX-Risk-Engine): A real-data FX risk service that aggregates multi-currency positions into a base currency and computes portfolio value, daily P&L, currency exposure, and ±10% what-if scenarios. - [My-Game-Theory-Lab](https://github.com/umutdinceryananer/My-Game-Theory-Lab): An Iterated Prisoner's Dilemma lab: configure payoff matrices and error rates, run tournaments between strategies, evolve genetic strategies, and inspect results with tables and heatmaps. — demo: https://umutdinceryananer.github.io/My-Game-Theory-Lab/ - [Mobile-Game-Analytics-Pipeline](https://github.com/umutdinceryananer/Mobile-Game-Analytics-Pipeline): A mobile-game user-acquisition analytics pipeline: SQL-driven KPIs, funnel + ROAS by channel, D1/D7 retention cohorts, and a churn model (Logistic Regression + XGBoost) on DuckDB, with a Tableau story. — demo: https://public.tableau.com/views/MobileGameUAStory/TableauStory (synthetic data) - [Petlyst-Web](https://github.com/PetlystHQ/Petlyst-Web): A team-built veterinary healthcare platform (CTIS senior project) connecting pet owners, vets, and clinics: a Node/Express + PostgreSQL backend with a TypeScript frontend, AWS S3 storage, token auth, and CI. - [Spotify-Playlist-Watcher](https://github.com/umutdinceryananer/Spotify-Playlist-Watcher): A scheduled service that watches public Spotify playlists for newly added tracks and sends a Telegram notification with an LLM-generated emotional/sentiment read of each track. Fully automated, zero-ops. - [Slack-Workflow-Engine](https://github.com/umutdinceryananer/Slack-Workflow-Engine): A config-driven Slack bot for multi-step approval workflows (refund / expense / PTO) with modals, multi-level approvers, database persistence, and structured logging. ## Open-source contributions - [elastic/kibana — PR #268326](https://github.com/elastic/kibana/pull/268326): Merged PR in elastic/kibana: show a "Go to dashboard" button in the save-success toast. (merged) - [langfuse/langfuse-docs — PR #2821](https://github.com/langfuse/langfuse-docs/pull/2821): PR in langfuse/langfuse-docs: add a Streamlit integration cookbook. (open) ## Private / in development - Hisar: A co-founded, in-development project that analyzes SEC filings to surface financial-risk signals. The working backend is an async pipeline — EDGAR ingestion → a rule-based gatekeeper (routine vs. material) → LLM scoring with multi-provider fallback (Anthropic primary, OpenAI/Google) → PostgreSQL → cross-filing pattern detection → "silence-first" notifications — and a separate research prototype adds no-look-ahead price alignment and a volatility-lift correlation method. Stack: Python, FastAPI, SQLModel/asyncpg, RabbitMQ, Redis, Docker, plus a FRED macro-rate pipeline. (not independently verifiable) - Themis: Offline-first iOS exam-prep app for Turkey's HMGS law exam, with a NestJS / Prisma / PostgreSQL backend and a React content backoffice, built on a two-way sync model (versioned content down, idempotent progress up). Solo-built and feature-complete through the study and mock-exam loop; StoreKit subscriptions are the remaining unbuilt piece. (not independently verifiable) ## What Umut is not good at yet - CUDA / GPU programming: Haven't trained on a GPU yet; zero hands-on so far, but genuinely keen to dive in. - Research depth: No research contribution yet; chasing at least a workshop paper. - Kubernetes: Deploy with Docker / Compose; no real k8s in production yet. - Advanced LLM internals: Shaky even on transformer internals; actively closing the gap. - Computer vision: Far from it, and honestly not drawn to it. - Rust / systems programming: Don't know it yet; keen to pick it up.