Jay
Prajapati

Backend Software Engineer building AI systems

Building @ Microchip Technology

Open to new opportunities

Bengaluru, India

GitHub955 contributions in the past year
LessMore

psst. this is my AI assistant.
it runs on a RAG engine I built. go on, grill it.

My actual profile, retrieved by Brain, the RAG engine I built. It also powers Offrloop and DocLens.

J.ai responses are AI generated, scroll the site to verify the details.

Under the hood

The chat up top isn't a wrapper around someone else's API. It runs on Brain, the RAG engine I built, on a server I run myself. Here's the path your question takes.

  1. 01

    You ask

    The question streams out of the chat box over server-sent events.

    Vue 3SSE

  2. 02

    Portfolio API

    Rate-limits, pins my full profile into the prompt, and logs the exchange.

    Node.jsExpressMongoDB

  3. 03

    Brain

    Rewrites follow-ups into a standalone query, then pulls the notes that matter.

    1. Embedbge-base-en-v1.5
    2. SearchQdrant · top 12
    3. Rerankcross-encoder · top 8

    PythonFastAPI

  4. 04

    Generate

    An open-weights model writes the answer as J.ai, grounded in those notes.

    gpt-oss:120bOllama Cloud

  5. 05

    Back to you

    Tokens stream straight back, with a filter scrubbing the model’s habits on the way.

    SSEstream filter

Everything except the model runs on one Hetzner box I provisioned and maintain: Docker, Caddy for TLS, and shared Postgres, Redis, Mongo and Qdrant serving seven apps. Images build in GitHub Actions and ship through GHCR; the server only pulls.

Who I am

I’m a software engineer who enjoys turning messy requirements into systems that feel simple to use and predictable to maintain. I work mostly with React, Python, RAG and a Database, and usually rely on my VPS for deployments and side projects to experiment with ideas outside work. Most of my learning comes from rebuilding things, questioning design decisions, and understanding why some systems stay effortless while others slowly collapse under their own complexity.

So far: 3+ years shipping production software, 3+ languages running in production, and 6+ systems live right now.

Not a designer. Not a manager. An engineer who ships.

Where I work

PythonFastAPIAzure OpenAIFAISSSentence TransformersMiniLM cross-encoderSQL ServerSSE

Built MAPS AI, a natural-language search engine for Microchip’s semiconductor catalog: about 60K parts, 806K parametric rows and 88M competitor cross-reference rows. Engineers type things like “automotive EEPROM, I2C, 256 Kbit or more” or “alternative to TI LMP2231” and get ranked parts with explanations. In the nine-stage pipeline the LLM only classifies intent and returns schema-validated JSON; it never writes SQL. Query terms are matched against three FAISS indexes (features, categories, brands) with rank fusion over embedding, lexical, alias and fuzzy scores, and an orchestrator fills predefined, parameterized SQL templates. When a search comes back empty, constraint relaxation drops the lowest-weight filters, and a MiniLM cross-encoder reranks using the user’s history. Every stage writes a trace with a paste-ready SQL query, and fixing data-layer issues took one query from zero results to 48 exact matches.

How it's built

  1. Parse queryspelling · units · terms
    Ontology3 FAISS indexes
  2. Classify intentAzure OpenAI · no SQL
  3. OrchestratorSQL templates
    CatalogSQL Server · 806K rows
  4. Relax & rerankMiniLM · user history
    Domain questionsinternal RAG
The LLM classifies. Deterministic SQL picks the parts.

Things I've built

By the way, my take on AI AI doesn't replace engineering, it multiplies it. It helps me ship scalable systems that millions use at work, and it lets me turn my own everyday problems into working solutions. I no longer wait for someone else to build the tool I need; if a problem bothers me enough, I build the fix myself. The AI speeds up the typing, while the architecture, judgment and care stay mine. A few of those builds are below.

Earlier:CloudNotes (2022), NewsBee (2022), TinyBasket (2020)

My stack, in sentences

Vue React TypeScript Python Java C# .NET Node.js FastAPI MongoDB Qdrant Redis SQL Server Flutter WebSockets

Build

Building backend services and APIs with C# and .NET (ASP.NET Core), Python (FastAPI), Java, and Node.js, usually as REST APIs and microservices shaped by Clean Architecture, with Vue 3, React, or React Native on top when the product needs a face.

Ship

Shipping with Docker and CI/CD (GitHub Actions, Azure DevOps) to AWS, Azure, and a server I run myself, backed by SQL Server, PostgreSQL, MongoDB, Redis, and Qdrant.

Operate

Focused on building systems that stay predictable as they grow through offline-first architecture, authentication flows, distributed workflows, and backend designs shaped around real-world usage instead of ideal-case assumptions.

Explore

Deep in AI engineering: AI agents with LangGraph and MCP, RAG and semantic search on Qdrant and FAISS, Azure OpenAI, and a daily workflow with Claude Code, Codex, and GitHub Copilot. Plus an M.Tech in AI/ML on the side.

Certificates

microsoft — Microsoft Certified: Azure Fundamentalsmeta — Programming with JavaScript by Metagoogle — Foundations of User Experience (UX) Design by Googlemeta — React Basics by Metameta — Advanced React by MetaAnthropic — Introduction to Model Context ProtocolAnthropic — Model Context Protocol: Advanced Topicsmeta — Meta Frontend Developer Professional CertificateMongoDB — Building AI-Powered Search with MongoDB Vector SearchMongoDB — Building RAG Apps Using MongoDB

Right now

Currently trying to balance full-time engineering work, AI/ML M.Tech coursework, and a growing obsession with AI systems, NLP workflows, and side projects that usually start as “small experiments” and somehow become full architectures a week later. Rebuilding things for no practical reason other than curiosity. In between, I still try to find time for books, badminton, and cricket before another idea decides it needs its own backend service!

Say hi

Hiring, building something with AI, or just want to argue about cricket? My inbox is open.

Jay
Email GitHub LinkedIn
Grab my résumé