AI engineer building agents that are observable, bounded, and grounded in real workflows.
MS Computer Science, University of Illinois Chicago (May 2026, GPA 3.89). I work on local-first AI, agent evaluation and observability, RAG, synthetic training data, and full-stack products in Python, Go and TypeScript. My fixes are merged into open-source AI tools like Haystack and OpenSRE.
Based in New Jersey · Open to full-time AI Engineer, ML Engineer and Software Engineer roles · OPT work authorization
Portfolio · Email · LinkedIn · Resume
Every project has a live, in-browser demo. No sign-up and no API key needed.
|
Local-first AI inbox triage. A scikit-learn + MiniLM classifier sorts mail in about 5 ms on CPU and calls a PydanticAI fallback only when it isn't confident. Corrections feed a guarded retraining loop, and every route comes with an explanation.
|
An operations console for AI agents. It records every run, exposes each step and tool call, and puts generated support drafts into a human review queue. External agents can send their own traces through an ingestion API.
|
|
Autonomous Kubernetes remediation with guardrails. A Go watchdog polls Prometheus, and a CrewAI orchestrator gathers evidence through four allowlisted MCP tools. It restarts on the first alert and doubles the memory limit only if the problem recurs.
|
Training data for tool-using agents. It turns ToolBench schemas into grounded multi-turn conversations: an endpoint graph, connected tool chains, schema-valid mock execution with session state, and an LLM judge with a repair loop.
|
I find real bugs in AI tools, reproduce them, and ship the fix with a regression test.
8 merged PRs across 7 projects. Rebuilt daily from the GitHub API (how).
| Project | Pull request | Merged |
|---|---|---|
| composio ★ 30.6k | Fix Pydantic field aliases in Python tool execution | Oct 2026 |
| baml ★ 9.4k | Teach initialize_runtime to handle trailing slashes | Oct 2026 |
| cli ★ 60 | Select ARMv6 binary on armv6l | Oct 2026 |
| txtai ★ 13k | Preserve ordered-list start values in HTMLToMarkdown | Oct 2026 |
| sentence-transformers ★ 19.2k | Rescore with the CrossEncoder when only min_score is set in mine_hard_negatives | Oct 2026 |
| txtai ★ 13k | Skip NULL aggregate values from shards with no matching rows | Oct 2026 |
| haystack ★ 26.8k | Treat None meta values as missing in MetaFieldRanker | Sep 2026 |
| opensre ★ 12.1k | Improve RCA prompt for evidence-grounded diagnosis | Apr 2026 |
Languages: Python, Go, TypeScript, JavaScript, SQL, Scala, C
AI and agents: PyTorch, Transformers, PydanticAI, CrewAI, LangGraph, LangChain, MCP, DSPy, RAG, Sentence Transformers, scikit-learn
Backend and data: FastAPI, Flask, Next.js, Postgres, SQLAlchemy, Drizzle, Pinecone, Pandas, NumPy, Spark
Infrastructure: Kubernetes, Docker, Prometheus, Cloudflare Workers, AWS, Azure, GitHub Actions, Linux
I'm interested in teams building AI agents, evaluation and observability infrastructure, developer tools and full-stack ML products. Reach me at gaurav.pvt25@gmail.com, on LinkedIn, or through my portfolio.