Ashutosh Verma

Full-stack + AI engineer. I build products on top of foundation models, from retrieval pipelines to the frontend people actually touch.

I'm a software developer with a year of professional experience and a slightly obsessive interest in the layer where full-stack engineering meets applied AI. Day to day I work in the usual web stack, but my real focus is building AI-native systems — retrieval-augmented generation, agent tooling, and the developer infrastructure around LLMs. I learn in public and treat every project as a chance to go one layer deeper than I strictly need to.

Work

Software Developer2025 — Present

[Company]

  • Developed and maintained features across a full-stack web application
  • Improved build reliability and reduced CI/CD deployment friction
  • Collaborated with product and QA on feature delivery and code reviews
  • Integrated AI tooling into development workflows

Projects

Production-style retrieval-augmented generation pipeline with hybrid search — BM25 + dense embeddings, Reciprocal Rank Fusion, ChromaDB, and a FastAPI service. Includes a RAGAS-inspired evaluation suite.

Python · FastAPI · ChromaDB · RAGAS

Cognitive-ability and reaction-time testing app with self-hosted collaborative editing. Built with a real-time collaborative document layer — full-stack product sense, frontend depth.

TypeScript · React · Next.js · Tiptap · HocusPocus

Collaborative editor with conflict-free sync built on Operational Transforms over WebSockets. Full unit test suite covering OT correctness and edge cases.

WebSockets · Operational Transforms

Custom MCP server that indexes codebases with tree-sitter so AI coding agents can search structurally instead of reading whole files — cutting token cost and improving accuracy.

TypeScript · tree-sitter · MCP

all projects

Writing

all writing