Building from India · working globally
Senior AI Engineer · Backend & Platform Lead|
I help AI products reach reliable production through architecture, backend engineering, and delivery.
Hiring for AI or backend leadership? View experience
Production record
About
My work spans AI systems, distributed backends, and developer infrastructure. I take products from architecture through implementation, deployment, and operation, so the important trade-offs stay connected.
I studied materials engineering at NIT Rourkela, then built my software depth through production systems, open source, and writing in public. That path made me comfortable working across boundaries, from a user problem and data model to code, cloud infrastructure, and on-call.
The work I care about starts after a demo. I build for real traffic, cost limits, failure modes, and handover. Evaluation, observability, rollback, and clear documentation are part of the product.
Each one written up in full — read the accounts
What I do
Useful AI, strong backend foundations, and systems designed to keep working long after the demo.
Production LLM engineering
Agent architectures, tool-execution frameworks, retrieval, evaluation, and voice interfaces — built to stay correct and observable after the demo.
Distributed systems at scale
Event-driven services, queues and workers, real-time transport, multi-tenant data models, auth and RBAC — designed for throughput and graceful failure.
Infrastructure and operations
Containerized deployments, infrastructure as code, CI/CD with safe rollback, observability, and production incident recovery.
Teams and architecture
Leading small engineering teams, owning architecture end to end, and staying hands-on across the stack while the system ships.
Operating model
The work is strongest when the decisions about product, architecture, delivery, and operations stay connected. This is how I take ownership across that path.
How I think
Opinions earned from systems that broke, budgets that ran out, and handovers that went badly. They shape every architecture I propose.
Most AI work fails after the demo, not during it. Retries, timeouts, evaluation, cost ceilings, and rollback are the product — not the polish you add later. I design for the second week of production, not the first hour.
Postgres, a queue, and a clear schema solve more problems than most architecture diagrams. I reach for novelty only where the problem is genuinely novel, because every extra moving part is a thing that pages someone at 3am.
Latency budgets, per-seat versus per-outcome pricing, and infra cost per request are design inputs. An architecture that ignores unit economics is an architecture that gets rewritten once the invoice arrives.
Problem to architecture to code to deploy to on-call. Handing a system off at the boundary is how context dies. The range is not a flex — it is what lets me make a decision on Monday that still holds in month six.
I publish the migrations, the incidents, and the things that broke. Writing forces the reasoning to be legible, and a post-mortem read by strangers is a stronger test of understanding than one read by nobody.
Fifteen engineers led or mentored. Shared schemas, review habits, and documented contracts outlast any service I write. If the system only works while I am watching it, I have not finished the job.
Measured outcomes
Results from shipped work. Client and employer names are deliberately omitted — the engineering is the point.
latency reduction
Re-architected LLM orchestration microservices sustaining 10k+ concurrent requests.
uptime sustained
Event-driven architecture over a message broker and cache for cross-service traffic.
deploy time
Automated CI/CD with infrastructure as code across production clusters.
faster agent integration
A real-time tool-execution framework that standardised how new AI agents ship.
faster ticket resolution
Real-time workflow automation on a multi-tenant field-operations platform.
daily alerts delivered
Resilient notification worker across SMS, email, and push with retry and failure triage.
The long game
From a non-CS degree in 2019 to leading AI engineering in 2026 — built by shipping things, not by credentials.
Enrolled in materials engineering — not computer science. Began teaching myself to program alongside the degree.
Paired client and server applications, written from scratch to find out how the halves of a real system actually fit together.
Moved past CRUD into engines, APIs, and developer tooling — the point where architecture, not syntax, became the interesting problem.
Completed the B.Tech + M.Tech track, moved into backend leadership, and published the first of 17 engineering articles.
Led teams of four and six. Took LLM workflows to 10k+ concurrent requests and cut deploys from 20 minutes to 3.
Owning architecture across AI orchestration, product backends, and cloud infrastructure. Ten articles published this year on AI systems and their economics.
Experience
From academic problem solving to backend leadership and AI products, each role expanded the scale of problems I could own.
Remote
Leading a five-person engineering team while building AI products, autonomous workflows, and the backend systems that support them.
Selected projects
Research, open source, and engineering practice — published in full. Client and product systems live under capabilities and impact instead.
Language AI · Research
A reproducible low-resource NLP pipeline evaluating IndicBART fine-tuning, retrieval baselines, and extractive heuristics for Maithili news.
Open source · Hospitality SaaS
A free hotel-operations platform for reservations, room and staff management, maintenance, analytics, and external booking coordination.
Open source · Engineering practice
A practical launch-readiness framework for founders and small teams shipping AI agents, workflow automation, search, and backend-heavy SaaS.
Technical range
The stack reflects what I actually use across AI, backend, data, cloud, infrastructure, and product engineering.
Work with me
Scoped engagements with a defined output. If your problem sits between these, say so — the shape is negotiable, the clarity is not.
Best for: teams about to launch
A structured review of an AI or backend system before it meets real traffic — reliability, APIs, auth, data, queues, CI/CD, and rollback — ending in a ranked fix plan rather than a list of opinions.
Best for: founders at zero
Turning a product idea into a system that can be built and afforded. Schema, service boundaries, build sequence, and the unit economics of running it — before anyone writes the first migration.
Best for: teams that need throughput
Embedded engineering across Go, Python, TypeScript, and cloud infrastructure. I ship features and also leave behind the review habits, shared contracts, and documentation that keep velocity after I leave.
Writing
Production stories, system migrations, backend patterns, and the failures that made the next version better.
AI · business · Jul 2026
AI · business · Jul 2026
AI · business · Jul 2026
AI · business · Jul 2026
AI · architecture · 9 reactions · Jun 2026
Feb 2026
go · GCP · 10 reactions · Jan 2026
supabase · postgres · 12 reactions · Jan 2026
Contact
I’m open to conversations around production AI, founding-engineer work, backend architecture, product partnerships, and focused system audits.