Hey, I'm Hardeep!

IIT Delhi → AI Engineer. Currently contracting at Squidgy AI (UK) and co-founding Coffee Coach. Building full-stack AI products with LLMs, RAG, and agentic systems.

Here's a quick intro about me && what I love to do

From the lab to the internet

I studied Chemical Engineering at IIT Delhi — where I learnt to think in systems and optimize under constraints. What I didn't expect was that the most interesting problems I'd encounter were messy, unstructured, and sitting inside data. That's what pulled me toward software.

Now I build AI products that take that mess and make it usable — pipelines, tooling, apps.

Hardeep in a cap and puffer jacket, outdoor selfie

Building things I actually want to exist

I like building things that solve problems for me, and seeing them become products that other people use. CoffeeCoach took off on Reddit without a single paid promotion. youtube-rag-scraper hit 76 GitHub stars because apparently a lot of people had the same frustration with video content being impossible to query. FluxRAG helped me understand RAG pipelines, eval and benchmarking which is enhancing my current AI Agent centric projects.

Hardeep standing on a mountaintop overlooking a valley

Life beyond the screen

I play chess obsessively. I played football every week until I tore my ACL last year. Still in recovery, still bitter about it. I befriend every mountain dog I meet. And I'm on an ongoing, probably never-ending hunt for the best chhole bhature in Delhi.

Hardeep's leg post ACL surgery, in recovery

Shipping is the habit

Since IIT Delhi I've been building independently — a COVID-19 crisis platform, two ongoing contracts (Squidgy AI and an NDA SaaS for a UK-based client), and full-stack AI products like Coffee Coach that people actually use.

I care about the things that matter in production — latency, evaluation, hallucination rates, cost per query. FluxRAG exists because I wanted to benchmark those tradeoffs properly before shipping anything.

Hardeep at a café in a pink polo, holding coffee with a bookshelf behind

My work history and education timeline.

4142 Ltd / Squidgy AI

2026 – Present

AI Software Engineer

Full-time contract at Squidgy AI, a multi-agent AI product by 4142 Ltd (UK). Working within the engineering team on AI-powered features and agentic systems, full-time remote mapped to UK working hours.

Stealth SaaS

2026 – Present

Sole Engineer (Freelance)

End-to-end ownership of a full-stack SaaS product for a UK-based client — architecture, build, and production deployment. Scope includes AI-powered features and agentic workflows. Client and product details under NDA.

Coffee Coach

2025 – Present

Founding Engineer

Built Coffee Coach 0 to 1 — a full-stack AI coaching app for specialty coffee brewers with personalised feedback via RAG pipelines and agentic feedback loops. 65 daily active users acquired organically via Twitter/X and Reddit with zero paid promotion.

Now co-founding and productizing with a product partner, building towards commercial launch. Stack: Next.js on Vercel, FastAPI on Hetzner VPS with Docker, Nginx, Certbot, PostgreSQL.

CovidWin

2021

Operations Lead

Co-built a COVID-19 resource platform from zero to full operations in 48 hours during India's second wave — 15 states, 50 cities, 10,000 verified life-saving resources.

Led volunteer coordination across 8 states, managing 4,000 volunteers from partner organisations. Built a 15-minute automated data sync pipeline using Google Sheets API with deduplication and multi-source aggregation.

Digital Marketing

2018 – 2024

Freelance

Independent client work across digital marketing — SEO, content, and campaign management. Later years included lightweight process automation, leading to a full pivot into software and AI engineering in 2025.

IIT Delhi

2013 – 2018

Chemical Engineering

Chemical Engineering at IIT Delhi providing the analytical foundation for systems thinking and constraints-based problem solving.

4142 Ltd / Squidgy AI

Full-time contract at Squidgy AI, a multi-agent AI product by 4142 Ltd (UK). Working within the engineering team on AI-powered features and agentic systems, full-time remote mapped to UK working hours.

Things I've built that people actually use.

coffeecoach.app

65 daily active users. Zero paid promotion.

Most coffee brewing advice online is scattered and contradictory. I ingested high-quality brewing data, made a RAG pipeline on top of it, and built an agentic coaching system based on that data — LLM orchestration with feedback loops that adapt recommendations based on user input.

Wrapped it all in a full-stack AI coaching app.

splitwala

Live in 117 groups. 2000+ active users.

Can't get your friends to install the bill splitting app? This Telegram bot makes it a single command — /split, /paid, /balances. Lives inside your Telegram group. No app to download, no account to create. Built on the principle that the best UX is the one that gets out of the way.

youtube-rag-scraper

76 GitHub stars. 20 forks.

YouTube has some of the best domain-specific knowledge on the internet, but it's locked in video format, impossible to query or reuse. This pipeline bulk-scrapes transcripts, processes them, and structures them into a searchable knowledge base for RAG systems. Handles chunking, embedding, and retrieval quality out of the box.

buildinpublic-x

$0.01/post on X. Zero friction. One time setup.

Most developers don't build in public because posting updates is friction. This simple GitHub Action reads your commit history over a few days, generates a thread via LLM, and posts to X and Bluesky automatically — no server, no backend, no SaaS. Lives entirely inside your repo.

flux-rag

Hybrid search + reranking. Async FastAPI server. 8 embedding models.

Most RAG systems ship without evaluation causing hallucinations to go undetected and retrieval quality degrading silently. FluxRAG inverts that: evaluation harness first, optimisation second. Drop in any file type across 10 formats, run a parameter sweep across chunking strategies and embedding models, and get a ranked benchmark report before you touch production.

Tracks latency, cost per query, and hallucination rate across every config because retrieval quality and cost tradeoffs shouldn't be assumptions.

GitHub contributions

1,418 contributions

GitHub Stars

87

Forks

21

Commits

330

Book a call with me

I'd love to chat even if there's no agenda!

August, 2026 — 30 min call

Toolbox

Check out my favorite tools and spots around the web.

Get in touch

I'm open to AI engineering roles, contract work, and interesting problems. If something I've built resonates, drop me a message!

  • Email hello@hardeep.cv
  • Address Based in Delhi, India
  • Hours Open to remote