- From:
- Shaswat Raj
- Looking-for:
- AI engineer, applied AI and LLM product internships or new-grad roles, at AI startups or AI teams in larger companies.
Subject: AI Engineer
I build LLM systems that run in production, and I fix bugs inside the agent tools other engineers use.
Two kinds of evidence. Upstream, my fixes are merged in Google's gemini-cli, T3 Code and the MoFA agent framework. In my own products, I run retrieval, MCP servers and agent-facing APIs that real users and real agents call.
Attachment: resume-ai.pdf · sh20raj@gmail.com · github
Evidence
Pressing Ctrl+C ended the turn but left the chat-compression model calls running. I threaded the turn's AbortSignal into compression in the client and in subagents.
Windows file-access events made the server announce a keybindings change on every revalidation. It now compares the resolved state structurally and only emits when something actually changed.
Protocol version negotiation, tool failures returned as readable results, every argument accepting what an agent already knows, and one core shared with the REST API and CLI. Runs on Cloudflare Workers.
Journals by writing, voice or video, answers with links to the entries used, voice calls with characters, connected apps with approval and an Android app. Early version, built solo end to end.
Replaced a panicking unwrap() in cron parsing with real errors, and made public enums #[non_exhaustive] so new variants stop being breaking changes.
Audits how visible a brand is to AI search engines and agents and turns the findings into a readiness score. CLI, FastAPI service and Streamlit app; team of two.
Stack
TypeScript · Python · Rust · OpenAI / Claude / Gemini APIs · RAG · Vector search · Model Context Protocol · AI SDK · FastAPI · Cloudflare Workers
How I work
- Treat the model's output as untrusted input, and design for the call that fails.
- Write tool responses a model can act on: what happened, and what to do next.
- Evaluate on real data before tuning prompts.
Other roles
- Software Development Engineer · I write code that passes review from people who don't know me.
- Forward Deployed Engineer · I ship in the user's environment, with their constraints, on their deadline.
- DevOps / Platform Engineer · I care about what happens after the merge: deploys, failure modes and observability.
- Full-Stack / Product Engineer · I take a product from an empty repo to people using it, and own every layer in between.
- Founding Engineer · I've built products from zero and run them alone, and I've led a team through a live event.