Open Source

Turns a GitHub issue into a pull request. Label an issue, and it reads it, plans which files need to change, writes the changes, and opens a PR against the default branch. Self-hosted FastAPI service plus a CLI, built on the Anthropic API and the GitHub REST API — reads the issue, plans the diff, generates full file content, and ships a branch + PR automatically.

Deploying an LLM behind vLLM + KServe on Kubernetes, with Kueue for GPU-job scheduling, Prometheus/Grafana observability, k6 load testing, and a FinOps cost model — the stack behind serving LLMs efficiently on shared GPU infrastructure. Runs on a laptop, no GPU or cloud bill required.

Multi-source job-alert bot on Telegram. Watches LinkedIn, Indeed, JustJoin.it, RemoteOK, Remotive, and We Work Remotely, then filters every hit by title, location, salary floor, and language before it reaches you. Reposts and duplicate roles are suppressed across company aliases, delivery can be instant or digested, and each alert comes with an in-chat Applied → Interview → Offer pipeline tracker.

Self-hosted ops agent on Telegram. Config-driven health checks (http, docker_container, file_contains, disk, command) run on an interval, locally or over SSH against remote servers via dedicated per-server keypairs. On failure it alerts on Telegram, waits out a configurable grace period for a human reply, and if none comes, hands an LLM a bash tool scoped to the affected machine — which diagnoses and fixes the issue using operator-written environment context plus the last 5 incidents as memory, then reports back and logs the outcome. A Telegram Mini App (Flask backend, HMAC-verified against Telegram's signed initData) handles check and server management, with a phone-and-email-verified admin approval gate for limited beta access.