Hi, I am Shivang. I live in San Francisco and work as a Founding Software Engineer at OHM, where we keep EV chargers and HVAC systems running across the country. I build the products that make that happen "Ohmie", our multi-agent AI; the mobile app our field technicians carry; and the platform that ties all operations together.
Joining a startup at day one means everything I ship starts as an empty repo and has to reach real users, so far that's apps on the App Store and Play Store, client-facing APIs, and AI systems in production. Before OHM, I built RAG-powered document intelligence at Steam Works Studio, growing it from a K-12 prototype to 200+ daily users across insurance, clinical, and legal.
Outside of work, I write about AI and engineering on Medium and live for hackathons, 10+ as a competitor, most recently on the other side of the table as a judge at AWS Builder Loft.
Fun fact: I am a multi-sport personality, any game going, count me in. Also on the roster: cooking, dancing, the gym, and a little DJing just for fun.
"Simplicity is prerequisite for reliability." -- Edsger W. Dijkstra
The short version of where I've been and what I shipped there. For the deep dives it's mainly architecture, features, and design decisions, open the Projects folder.
Core member of the founding team, from empty repos to products in the field:
Led AI document intelligence from prototype to 200+ daily users:
Research engineering on real clinical data:
Where money literally flowed through my code:
The AI agent at the heart of OHM. Ask it anything about operational data, payments, invoices, service history, client records, in plain English, and it orchestrates the right agents with tool calling, knowledge extraction, and fuzzy search to answer accurately. Extended across the platform: worktype building, service diagnosis, scheduling next visits, and parsing documents into service knowledge.
The operations command center: scheduling, monitoring, and tracking every field service job for EV chargers and HVAC systems across customers. Built with Next.js, React, and Django, powered by Ohmie, with TanStack Query caching and context management that cut response latency from 400ms to under 10ms.
React Native app shipped on the App Store and Play Store, used daily by field technicians for on-site inspections and servicing. Mapbox-powered unit maps, guided work flows, and real-time diagnostic capture that feeds straight into customer-facing reporting dashboards.
How enterprise clients talk to OHM: standardized REST endpoints for station imports, inspection reporting, work order management, inventory tracking, and technician scheduling, including a full ChargePoint integration and a client-agnostic public API serving 10,000+ work orders.
Full-stack document intelligence with Django & LangChain: parsing pipeline (Reducto, LlamaParse) over 1,000+ pages, OpenAI embeddings, and vector search powering natural-language Q&A. Started with K-12 curriculum, grew into insurance, clinical, and legal, 200+ daily users at sub-200ms.
Text a photo of your grocery receipt; Sous stocks a virtual pantry with AI-estimated expiry dates, suggests recipes that use expiring items first, and replies with AI-generated videos of the dish. FastAPI backend, vision models, iMessage transport, and persistent memory for allergies and past meals.
Upload your ChatGPT or Claude conversation exports and see your thinking as interactive bubble charts, topics, questioning styles, AI personality patterns. Everything runs in the browser: Transformers.js embeddings, WebLLM running Phi-3-mini locally, K-means++ clustering, D3 visualization. Zero servers, total privacy.
A GPT built from first principles in PyTorch: neural net primitives, backprop, attention, and the full transformer stack, plus a BPE tokenizer, KV-cache for fast inference, and grouped query attention. Because the best way to understand LLMs is to build one.
React Native app that uses the Gray Whale recommendation algorithm to suggest outfits you'll actually want to wear. A personal stylist in your pocket, built against the clock with a team.
Interactive leaderboard visualizations of treatment outcomes built on University of Chicago datasets at UMass Boston's MPsych Lab, letting clinicians compare protocol efficacy across 5,000+ patient records.
Automated verification of cash and coin flows across 10,000+ devices in casinos and vending machines nationwide, $2M+ in daily transactions validated at 99.9% accuracy, with JUnit suites automated through AWS Lambda workflows.
Multi-sport personality: if there's a game going, count me in. Any field, any court, any team that needs one more.
Strictly a party trick, but hand me the decks and I'll keep the floor moving. Reading a crowd is half the fun.
I cook the way I build software: start with a recipe, trust the taste tests, ship the dish. Experiments encouraged.
Committed to the gym, lifting keeps me sane and disciplined. A healthy body powers a sharp mind.
Yes, engineers dance. Music moves through DJing into dancing, it's all the same energy, just different outputs.
Master of Science in Computer Science • May 2023
Boston, Massachusetts
Bachelor of Engineering in Computer Engineering • May 2021
Mumbai, India
Co-authored paper published at the 2020 Second International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, under the Machine Learning domain.
Demonstrated rapid prototyping and cross-functional collaboration across 10+ hackathons.
Judged 40+ projects in the agentic AI and machine learning space, shortlisting 5 winning teams. Sponsored by HydraDB, Photon, RocketRide, and more.
Developed 4+ full-stack SaaS products using TypeScript, NextJS, MaterialUI, Stripe API, LLMs, and Pinecone Vector DB, leading a team of 3 developers.
My writings on Medium, View all posts on Medium ↗
October 10, 2026 • Towards AI
I went back to the 2017 paper behind every AI model to see how much of the original is left. Almost nothing, and the two parts that survived explain why your chatbot rarely gives the same answer twice.
September 24, 2026 • Towards AI
A punctuation mark turned into a giveaway, users complained, and the newest models learned to hide it. What 490,000 webpages, 69,000 medical preprints, and a cross-model study say about where AI writing style actually comes from.
September 13, 2026 • Bootcamp
The story behind this very site: draggable XP windows, Minesweeper, a chatbot that isn't AI on purpose, Breakout.exe, and how it's hosted for free. You're looking at it right now.
June 12, 2026 • Towards AI
I tested four Claude models on real-world tasks, no benchmark datasets, just actual work, and the results didn't follow the leaderboards.
May 1, 2026
Your docs aren't just for humans anymore, they're the context your AI tools run on. Why writing things down became a first-class engineering skill.
April 16, 2026 • Towards AI
Production AI doesn't have to start expensive. How far a free tier can actually go when you architect around its limits.
April 11, 2026 • Towards AI
The gap between an agent demo and an agent your users rely on is enormous. Lessons from shipping agentic systems for real operations.
April 9, 2026
Sometimes the most productive engineering session produces no code at all. On the underrated craft of thinking before typing.
December 23, 2025
AI writes a lot of code now, but architecture, scaling decisions, and knowing what to build are still on you.
December 16, 2025
When AI hit the gym too hard: why smaller, specialized models are quietly becoming the most interesting story in the field.
September 4, 2025
The unglamorous infrastructure that quietly powers everything, and what it tells us about where AI is headed.
December 4, 2024
A tour through the AI agent ecosystem, what agents actually are, the leading frameworks, and how to pick between them.
December 4, 2024
A technical assessment turned into a full Streamlit web app, upload flow log files and lookup tables, process data, and visualize results in real time.
July 1, 2020 • Analytics Vidhya
Built a local web app using the Python wordcloud library and Flask for creating wordclouds that even unfamiliar users could use. Published in Analytics Vidhya.
June 11, 2020
A reflection on how the human mind communicates through five senses, creating impulses of thoughts, and how surrounding yourself with good influences and meditation can shape your path.
Hey! I'm Shivang's digital twin. Ask me anything about my experience, tech stack, what I'm building, or why I think tabs are superior to spaces. Pick a question below or type your own!
Left-click: reveal • Right-click (or 🚩 mode): flag
Click a card, then click where to move it. Double-click sends it to the foundations. Click the deck to draw.
Shivang Raikar