Skip to content
View B0yko's full-sized avatar
3 followers · 0 following

Achievements

Achievement: Starstruck

Achievements

Achievement: Starstruck

Highlights

  • Pro

Block or report B0yko

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
B0yko/README.md

Hi, I'm Andrii Boiko.

I'm a Lead AI Engineer based in Berlin. I work on production AI architecture, agents, retrieval, and evaluation.

Case studies & personal projects · LinkedIn · Hugging Face · Email

Selected projects

Project What it does
pgwarden An MCP gateway for Postgres with per-person database permissions, PII masking, human-approved writes, and auditable queries.
taskdistill Distils recurring LLM tasks into small local models with LoRA and MLX, held-out evaluation, and selective escalation to the original API.
agent-claimcheck Checks agent success claims against tool receipts and state evidence. Includes calibrated decisions, a review queue, and a labelled synthetic benchmark on Hugging Face.
Tern Local multimodal search for video and podcast archives on Apple Silicon, combining speech, on-screen text, and visual retrieval.
local-enough Measures local and cloud models on business tasks: quality, latency, cost, and routing decisions against explicit quality thresholds.
booking-truth Tests AI booking agents on the calendar state they leave behind, with fault injection, pass^k, and false-success rates. Ships a guarded Cal.com, Google Calendar, and HubSpot booking agent.

These are personal projects. Each repository documents its implementation, evaluation, and current limitations. Tern is source-visible under an all-rights-reserved licence; the other projects above use Apache-2.0.

Production work

My case studies explain the architecture and engineering decisions behind that work.

Tools I work with

Python · TypeScript · FastAPI · LangGraph · LangChain · PostgreSQL / pgvector · MCP · Docker

Pinned Loading

  1. tern tern Public

    Local AI search for podcast and video archives on Apple Silicon: Whisper speech, Apple Vision OCR and SigLIP-2 visual search fused into one ranked list, with a trim editor and FCPXML export. FastAP…

    Python 42

  2. taskdistill taskdistill Public

    Distil an expensive LLM API call on a narrow task into a small local model: capture traffic, curate, LoRA fine-tune with MLX on Apple Silicon, evaluate against the teacher, and serve an OpenAI-comp…

    Python 3

  3. local-enough local-enough Public

    Measure whether a back-office AI task can run on your own hardware: accuracy vs gold, p50/p95 latency, $/1k tasks, amortised hardware cost and break-even volume, plus an OpenAI-compatible router th…

    Python

  4. pgwarden pgwarden Public

    Governed Postgres access for AI assistants: an MCP gateway with OAuth 2.1 and OIDC login, one Postgres role per person, database-enforced read-only queries, PII masking, human-approved writes, a ta…

    Python

  5. agent-claimcheck agent-claimcheck Public

    Catch AI agents that say "done" when the task isn't. Checks agent-trace/v1 traces with claim-evidence rules, LLM judges and a trained classifier, returns verified / false_success / unverifiable wit…

    Python

  6. booking-truth booking-truth Public

    Tests any AI appointment-setting agent on the calendar's end state, not its transcript: fault injection, pass^k, false-success rate. Ships a guarded Cal.com / Google Calendar / HubSpot booking agent.

    Python