Build an AI product
Take a focused product from idea to production: user experience, APIs, agent workflows, tool calling, memory, evaluations, and the infrastructure around it.
Best for a new product, feature, or tightly scoped MVP.
Available for select projects
I turn ambitious AI ideas into reliable software.
I help startups design, build, and harden AI products—from the interface and APIs to agents, integrations, infrastructure, and everything production demands.
Now Founding engineer at Curvo AI
Before Founding engineer at MonkCI
Focus AI systems built for production
Base India · working globally
01 / Selected work
Complex systems, made dependable.
A closer look at the architecture, tradeoffs, and engineering judgment behind the work. Professional details are kept public-safe.
01Production AI at the CRM boundary
AI features fail in production when context arrives late, duplicated, or incomplete — and the model still answers with confidence.
02Industrial agent memory as infrastructureAn industrial agent that cannot remember what failed, what was tried, and why a decision was made will repeat expensive mistakes.
03Reproducible debugging and harness engineeringIrreproducible production failures are expensive twice: once to understand, and again every time an agent or teammate re-derives the same diagnosis.
04CI workers, queues, and distributed executionDistributed execution is easy to sketch and hard to operate: every retry, timeout, and lost event becomes someone else’s broken build.
02 / Work with me
Senior execution without the hiring cycle.
Focused engagements for startups that need an experienced builder to own a meaningful technical outcome.
Integrate and automate
Connect APIs, CRMs, webhooks, and internal workflows with deliberate handling for identity, retries, background jobs, and partial failures.
Best when systems need to exchange real, messy data reliably.
Stabilize and scale
Turn a fragile prototype or difficult production issue into a dependable system through architecture review, observability, reproducible debugging, and hardening.
Best when the demo works but production does not.
Have a different problem? The best projects rarely fit a template.
Tell me about it03 / How I work
Built with production in mind from day one.
Fast iteration matters. So do the details that keep software useful after launch: clear boundaries, observability, and reproducible failures.
Production over demos
A system is not useful because it works once in a notebook or a staging screenshot. It is useful when it survives retries, partial data, bad clocks, and the next deploy. Demo-quality AI is a starting point, not a finish line.
Context is infrastructure
AI systems depend on reliable context: what is stored, what is retrieved, what is excluded, and who is allowed to see it. Memory, boundaries, and retrieval contracts are infrastructure decisions. Treat them with the same seriousness as a database schema.
Make failures reproducible
The first occurrence of a difficult failure requires investigation. The second should be cheap. A captured, replayable failure is dramatically easier for humans and coding agents to diagnose than a pile of correlating logs.
Prefer simple orchestration
Do not add coordinators, multi-agent meshes, or extra queues unless they create leverage you can name. Extra coordination hides the data flow and multiplies failure modes. Earn complexity.
Let’s build something useful
Bring me the ambitious idea or the stubborn problem.
I’ll help turn it into a clear plan and dependable software.