I build AI products that stay correct.

Full-stack and GenAI engineer in India. TypeScript, Bun, Hono, Postgres, Qdrant. Former Head of Technology at Senslyze.

Read the writing See the work

Two requests, one key

AGET user:123idle

BGET user:123idle

ASET processingaccepted

BSET processingaccepted

Both writes pass. Two jobs now run for the same user.

Same key, same millisecond. Redis runs each command in order. It does not run your check and your write as one thing.

Selected work

Real-estate WhatsApp platformLLM platform, document Q&ASenslyze Government RAG backendRetrieval, launch deadlineSenslyze Client reports from Git historyAgents, human reviewSenslyze COBOL to Java migrationMulti-agent prototypePersistent

WhatsApp, property Q&Alive

“3BHK near Manish Nagar, ready to move?”

Two listings match. One has a 40 ft frontage and a covered parking spot.

Answer built from

brochure.pdf, p.4price-list.xlsxsite-plan.jpg

Site visitSat, 11:00

Created byexplicit backend code

RAG, retrieval tracequery 481

Querynotice period for renewal

lease-clause.pdf0.82

policy-2026.md0.74

faq.json0.61

Top chunk says 60 days. The answer cites that chunk and nothing else.

Git history to client reportOctober

feat: add bulk export12 Oct

fix: retry on timeout18 Oct

chore: bump deps24 Oct

Shipped bulk export. Fixed a timeout that blocked large files. Dependency work stayed internal.

Reviewed byPranav

COBOL to Java, agent looprun 12

Coder proposes the portdone

Reviewer checks the diffdone

Human approves the edge casewaiting

Open questionrounding on interest

About

I started in backends and stayed there: APIs, databases, integrations, deployments. For a while I led technology at Senslyze, building LLM products end to end with a team of eight to ten developers.

Most of my work now is agentic AI: retrieval, tool calls, and the backend code that keeps every action correct. I care more about what happens when a step fails than about the demo.

I learn by pushing back. When an AI tells me a system is efficient, I ask: compared to what? More than once, that question turned a confident answer into a correction.

I use coding agents every day and build the guardrails around them: repository specs, lint boundaries, and checks that fail loudly.

Pranav Bobde

TypeScript engineer, Nagpur

Writing

I write to find out what I actually believe.

The bug Redis didn't causeTwo requests, one key, and a rabbit hole through Durable Objects, actors, sagas, and workflow engines. The fix was one command.03 Nov 20269 min readRead the post

Longer notes on systems and AI products. New posts go out with the newsletter.

I'm open to hands-on full-stack and GenAI work, and to contract work on LLM products.

bobdep31@gmail.com LinkedIn GitHub

Nagpur, India. Open to remote or Bangalore.