Neel Raval

I build products people didn’t know they needed.

I’ve been building software since I was in 8th grade - started with Python, mostly self-taught, picked up a B.Tech along the way. These days I’m architecting an AI-native EMR from the ground up, and a separate platform that automates the painful manual work of public-sector tendering. The through-line across both - and across most of what I build outside work - is UX. I believe the difference between software that gets adopted and software that gets abandoned is almost always in the interface, in the accessibility, in the small details that make a system feel trustworthy and worth a person’s time. I work at the intersection of product, engineering, and design, where craft compounds.

Currently: Architecting an AI-native EMR - clinical workflows, agentic interfaces, the whole stack.
Outside of work: Gym, bowling, cooking, eating my way through Bengaluru, planning the next trip, or down a YouTube rabbit hole.
Based in: Bengaluru, India

Experience

Where I’ve built.

Assistant Zonal Head, South

Viral Fission Remote Jan 2021 – Apr 2023

Ambassador Team Lead City Lead Student Manager Assistant Zonal Head
- Worked with ~20 brands across tech, FMCG, streaming and services on brand amplification, sampling surveys, and content creation, hitting 90% brand satisfaction.
- Coordinated a network of 6000+ ambassadors and ran 20+ offline events to drive brand engagement.
MERN Stack Intern

Bizco Remote Jul 2022 – Sep 2022
- Built an e-commerce Kirana marketplace on the MERN stack to bring neighborhood retail online.
- Implemented offline-first payments (Paystack) and offline data sync (IBM Cloudant + PouchDB) for reliable use under intermittent connectivity.
- Shipped REST API and frontend with CI/CD on Heroku and Netlify.
Junior Developer

Voltfraction Remote Oct 2020 – Jan 2021
- Built an internal CRM to streamline company operations, with a Django backend and SQL data layer.

FLEXIUS

Flagship

An AI-coached fitness PWA that turns workout tracking into a conversation.

I wanted workout logging to feel like talking to a coach, not filling out a form. You speak or type - "did 3 sets of bench at 65kg", "plan a push day" - and the model decides whether to log a set, pull your history, generate a structured plan, or just answer a fitness question. It's offline-first, so it keeps logging through dead gym wifi via an IndexedDB queue and replays when you reconnect. The piece I'm proudest of is the provider-neutral LLM layer - Gemini, Claude, and OpenAI all run behind one interface, with bring-your-own-key support.

Next.js 16
React 19
Turso (LibSQL)
Drizzle ORM
Tailwind 4
shadcn/ui
Web Push
Vercel

Live GitHub

Jogen

Honest resume tailoring advice - line edits and skill-gap flags, never fabrication.

Most resume tools optimize for "rewrite this for me." I built Jogen to do the opposite - it's an advisor, not a ghostwriter. Upload a resume and a job description and you get a fit score, verbatim line edits you can paste anywhere, and skill gaps each paired with a concrete first step. It never invents experience, and a Pydantic contract gate rejects any model response that doesn't meet the schema.

React 19
TypeScript
Vite
FastAPI
Pydantic
Anthropic SDK
SSE
Render

Live GitHub

Instagram Reel Analyzer

A production-shaped vision-LLM API for analyzing Instagram reels.

Started as a quick hack - share a reel from iOS, get a structured analysis back - and grew into a proper multi-tenant service. Provider-abstracted vision inference runs against Gemini or a local Qwen2.5-VL, with Redis-backed API keys hashed at rest, two-window rate limiting, and a schema-versioned cache. Structured outputs are enforced at decode time via Pydantic, not parsed after the fact.

Python 3.13
FastAPI
Pydantic v2
Redis
yt-dlp
Tenacity
structlog
Docker

GitHub

Sanjeevani

Offline clinical decision support for rural primary health centers.

A doctor enters a patient presentation and an optional photo, and six chained LLM calls work through feature extraction, differentials, triage, investigations, referral, and an action plan - ending in a patient summary in their own language. It runs fully offline on Gemma 4 via Ollama, so no patient data leaves the device, and investigations are bounded to what a real PHC can actually do.

Python
Gemma 4 (Ollama)
FastAPI
Gradio
SSE
Pydantic

GitHub

SOC Analyst

An RL-style environment for training and evaluating LLM agents on security operations work.

A stateful simulator of three SOC analyst tasks of increasing difficulty - triage, investigation, and correlating a ten-alert coordinated attack campaign out of a noise floor. Agents interact over HTTP with structured actions, scored by task-specific graders plus dense reward shaping with the asymmetric penalties a real SOC operates under. Built on Meta's OpenEnv framework.

Python
FastAPI
Pydantic
OpenEnv
Docker
HF Spaces

GitHub HF Space

PostgreSQL

Product Strategy

Git

Agile / Scrum

Turso

WebSockets

Engineering Leadership

Google Gemini

LLMs

Next.js

Structured Logging (structlog)

Vercel

Django

Streaming Inference

Server-Sent Events (SSE)

Docker

Node.js

Render

Python

gRPC

Redis

Agentic LLM Systems

FastAPI

Technical Roadmapping

GitHub Actions

HuggingFace Spaces

Retrieval-Augmented Generation (RAG)

Mentorship

React Native

OpenAI

Rate Limiting

Multimodal AI

Embeddings

Convex

React

MongoDB

Computer Vision

Testing (Vitest, Jest, Mocha, Chai)

TypeScript

System Architecture

Nginx

GraphQL

Tool-Use & Function Calling

AWS

REST APIs

Express

Code Review

Drizzle

Anthropic

Pydantic

Prompt Engineering

Zustand

Prisma

LLM Evaluation & Reward Shaping

JavaScript

TailwindCSS

CI/CD

Anthropic Claude

Streaming

tRPC

Local LLM Deployment

Structured Outputs

Progressive Web Apps (PWA)

Authentication (Auth0, OAuth)

Found 0 / 64

I build products people didn’t know they needed.

Where I’ve built.

Before engineering was the whole job - dev roles, and the brand-side work that taught me how products actually reach people.

Assistant Zonal Head, South

MERN Stack Intern

Junior Developer