Harsh Kedia, assembled.
I'm an AI fullstack engineer. I build production systems end-to-end, from LLM and agent infrastructure to the backend and the frontend on top. Most recently at shoppin', where I built an engine that turns a garment photo into a factory-ready sewing pattern in minutes, plus the recommendation feed, the payments stack, and the try-on around it.
I got here sideways. I wanted to be an astrophysicist; that didn't work out. But I'd always been the kid taking the machine apart, modding games, rooting phones, flashing custom ROMs, and when I realized a web page was just something you could make, that was it. HTML pulled me in, Python kept me, and somewhere in there it stopped being a backup plan and became the thing I love.
I chase hard problems: the ones that supposedly can't be done, or can't be done cleanly. That's where the kick is, even when it means obsessing over something and not sleeping until it gives. I'm unusually good at two things: debugging (finding the why and the where when nothing else will), and taking a project from an empty repo to the best version of itself.
I chose Shoppin to go deep in a domain I didn't know, and fashion delivered: how the vertical works, how the supply chain moves, and now, with custom garments, how you compress days of expert pattern-making into minutes. Off the clock I'm watching football, playing cricket, and, inexplicably, getting into pickleball. I'm currently between roles and looking: open to senior and founding-engineer seats where I own hard systems end-to-end.
- → Built an agent-based pattern-making engine: garment image/spec → production-ready ASTM sewing pattern in minutes, ~95% accuracy
- → Shipped an AI-native custom-clothing platform end-to-end: talk it out, design in-browser, get real made-to-order clothes
- → Owned the recommendation & discovery engine (250K+ items) and the entire e-commerce & payments stack
- → Built the entire backend from scratch for a multi-modal agentic AI platform (Chat / Email / WhatsApp / Voice)
- → Hybrid RAG pipeline with Milvus: sub-200ms retrieval, 95%+ context accuracy over multi-turn conversations
- → Shipped multi-provider LLM infra (OpenAI, Anthropic, LLaMA, Qwen) via gRPC, adopted as the standard across every team
- → Built a 24/7 autonomous freight-forwarding agent end-to-end for a logistics startup
- → Reads inbound emails, understands intent with an LLM, and extracts structure from messy shipping docs
- → Agentic quoting loop that fans out to partner agents and returns the best option with no human in the loop
- → Secure file-sharing platform on Azure: 6+ microservices, SAML SSO, OAuth, 2FA, DLP, malware scanning
- → Enterprise clients: Toyota and UCO Bank
- → Backend rewrite: 130+ REST APIs; 1s → 30ms latency, 2s → 0.4s page loads
The best software feels like nothing at all.
It gets out of the way, it respects your time, and it only shows up when you need it to.