Switch to English?
Yes
Переключитись на українську?
Так
Переключиться на русскую?
Да
Przełączyć się na polską?
Tak

Dmytro Staroselskyi

Offer Dmytro work on your next project.

Ukraine Lvov, Ukraine
18 hours 53 minutes back
Available for hire available for hire
on the service 6 years

Rating

Successful projects
No data
Average rating
No data
Rating
198
AI & Machine Learning
473 place out of 2887
Bot Development 1
726 place out of 1946

Language proficiency level

Українська Українська: fluent
English English: upper-intermediate

Skills and abilities

Portfolio


  • Eval-Lab — Regression of prompts and models

    AI & Machine Learning
    Web dashboard for regression testing of prompts and models. Running a test set through two models/prompts — comparison based on 4 sub-scores.

    Technically interesting aspects:
    — LLM-as-judge through 5 providers (OpenRouter, Anthropic via tool-use, Gemini, Groq, mock)
    — 4 sub-scores for each case: correctness, relevance, completeness, prompt_quality
    — Cap on final score for poor prompt — prevents strong model from masking poor prompt
    — Per-provider throttle and retry with backoff + Retry-After
    — Mock mode for running without API keys (CI-friendly, $0)
    — Editing secrets in logs

    Stack: FastAPI, async SQLAlchemy, Alembic, httpx, Pydantic, vanilla JS, Docker.
  • AI Research Agent on LangGraph

    AI & Machine Learning
    An agent on LangGraph that autonomously plans searches, simultaneously browsing the web and arXiv, synthesizes responses with sources.

    Technically interesting aspects:
    — Parallel fan-out/fan-in graph with a reducer on tool_errors (the graph does not crash on provider failure)
    — 4-provider search fallback: Tavily → SearchAPI → DuckDuckGo → Wikipedia
    — Source-grounded synthesis: refuses to respond without sources (anti-hallucination)
    — PostgreSQL persistence with SQLAlchemy + Alembic (4 migrations)
    — Robust JSON parsing with code-fence stripping and regex fallback

    Stack: LangGraph, FastAPI, PostgreSQL, SQLAlchemy, Alembic, OpenRouter, Tavily, BeautifulSoup, Docker.
  • AI Radar — RAG Research Assistant

    Bot Development
    RAG-bot that collects AI news from 7 sources and answers questions with links to the sources.

    What is interesting technically:
    — Hybrid search: BM25 + vector search Qdrant, merged through Reciprocal Rank Fusion, plus cross-encoder reranker
    — RAGAS evaluation of answer quality (faithfulness, answer_relevancy)
    — Auto-indexing via APScheduler every 6 hours
    — Telegram bot as a thin httpx client to FastAPI

    Stack: Python, FastAPI, LangChain, Qdrant, rank_bm25, sentence-transformers, aiogram, APScheduler, Docker, GitHub Actions.