Saumya Jain
GenAI & Full Stack Engineer
Building agentic RAG systems, production APIs, and web platforms that behave predictably against messy real-world data.
{ "name": "Saumya Jain", "role": "GenAI & Full Stack Engineer", "genAI": [ "Agentic RAG", "LLM Tool-Calling", "Evals" ], "stack": [ "TypeScript", "Next.js", "NestJS", "Postgres" ], "passion": "Agentic RAG Systems" }
About Me
I'm Saumya Jain, a Full Stack Engineer with 4+ years building production web systems, now focused on GenAI application engineering. I design and ship agentic RAG systems end to end: tool-calling LLM orchestration, hybrid vector and keyword retrieval on pgvector, asynchronous document ingestion, and automated evaluation harnesses enforced as CI regression gates. Underneath that sits a deep TypeScript and Node.js foundation — React/Next.js frontends and NestJS microservices, with PostgreSQL, MongoDB, Redis, Docker, and AWS in production. My focus is making probabilistic systems behave predictably against messy real-world data, and proving it with measured accuracy rather than claims.
Technical Arsenal
AI & GenAI
Languages
Frontend
Backend & APIs
Databases & Data
Cloud & DevOps
Testing & Tools
Work Experience
Software Engineer – Full Stack & E-Commerce
Izna
- Owned end-to-end order lifecycle management for voscollection.com — cart creation, checkout, order placement, inventory decrement, status transitions, and post-order updates — ensuring data consistency across the flow.
- Designed and implemented RESTful order, cart, and inventory APIs, handling edge cases such as stock conflicts, partial fulfilment, and status rollbacks to keep order state accurate at scale.
- Built Vestum, a family wealth-management platform on Next.js, with multi-provider authentication (Google OAuth, email/password, and biometric/passkey sign-in with persistent device sessions), a live market-data panel, and Sentry instrumentation in production.
- Built and maintained voscollection.com, a headless CMS-driven site for a North American medical-aesthetics network, using Next.js and Sanity CMS for structured content pipelines and real-time updates across clinic and product pages.
Software Engineer – Full Stack & Backend
Wisflux Tech Labs
- Built and maintained a microservices architecture using NestJS and an NX monorepo, improving API response time by 25% and reducing server load by 30%.
- Optimised PostgreSQL queries and indexing strategies, reducing page load times by 35%, and implemented Redis caching layers to improve read performance across high-traffic endpoints.
- Developed full-stack features using React.js, Node.js, and PostgreSQL across multiple client-facing web applications, collaborating with cross-functional teams through delivery.
- Engineered an automation module that significantly reduced manual operations and increased overall system efficiency.
Software Engineer Intern
Celebgaze Media
- Translated Figma designs into pixel-perfect, cross-platform Flutter screens.
- Integrated REST APIs for authentication flows, dynamic content delivery, and real-time data.
Featured Projects
Writing
Notes on AI application engineering, backend architecture, and the parts of production engineering that only show up in year two.
The RAG System With No Embeddings
An embedding pipeline is not a decision you make once — it is a background system you own forever. Here is what retrieval looks like when you refuse to build one.
Retrieval Should Be a Tool, Not a Pipeline Stage
A fixed pipeline retrieves before every answer, including the ones that need arithmetic, and the ones that need nothing. Making retrieval a tool lets the model decide — and lets it look twice.