I'm a software engineer who genuinely enjoys building things. Not just shipping features — understanding systems end to end, from the database query plan to the user experience. Five years across fintech, gaming, and enterprise AI, and the pattern is the same: I see a problem, I build something to solve it, and I learn something new along the way.
At Walmart, I build AI platforms — production RAG systems, multi-agent orchestration, and LLM-powered developer tooling used across the org. Before that, I was the founding engineer at Podeum, taking a gaming platform from zero to 10K daily users. The common thread: I gravitate toward hard problems where distributed systems meet emerging technology.
When a new tool drops — a new LLM, an agent framework, a dev tool — I don't just read the blog post. I build something with it. That's where the experiments below come from. Some turned into product concepts, some into architecture patterns I now use at work, and some were just fun to build.
IIT (B.H.U), Varanasi
B.Tech (2016–2020)5+ Years
Fintech · Gaming · Enterprise AIBuilding AI Platforms
At WalmartOn My Workbench Right Now
RAG pipeline with hybrid retrieval (BM25 + semantic, RRF, cross-encoder re-ranking), Vision LLMs for OCR, and LLM response generation with evaluation. Live in production at Walmart.
Stateful agent coordination with task decomposition, parallel execution, and deterministic pause/resume. Powers human-in-the-loop code generation and review cycles in production.
1500+ synchronous executions/sec. P95 latency cut from 200ms to 130ms. Multi-tenant architecture isolating multiple teams and workflow engines on a shared cluster.
Built the entire backend from scratch as founding engineer — auth, virtual economy, in-app purchases, real-time live scores. Scaled to 10K daily users.
High-volume batch processing across multiple Rakuten brands. Built a mock server framework to simulate external payment providers for reliable integration testing.
Spec-driven development workflows using LLMs. Internal tools that make the whole backend team faster — not just me. Introduced AI-assisted code review and generation cycles.
Things I build when a new idea won't leave me alone
What if you could debug a codebase like you debug code? Breakpoints, stepping, call stacks, and "pause and ask why" — applied to static code exploration. Full PRD, competitive analysis of 13 tools, and an interactive mock.
An offline-first travel planning app for friend groups. Full-stack: Spring Boot backend, Expo/React Native mobile, PostgreSQL. Sixteen trip sections, section-level privacy, day-by-day timeline. Spec-driven from day one.
An AI agent framework with persistent memory, skill management, multi-platform connectivity, and autonomous task execution. Built to explore what happens when agents can learn from their own mistakes and improve over time.
Can you separate the brain that plans from the hands that act? This architecture lets you pick any advisor model, any executor model, and any MCP-compatible toolset. Runnable Python kernel, architecture guide, and do's/don'ts.
Always up for talking about systems design, AI infrastructure, or whatever new tool came out this week. If you're building something interesting, I'd love to hear about it.