Backend and microservices
Service boundaries, typed APIs, and data models that still make sense after the product changes shape.
- Python
- FastAPI
- PostgreSQL
backend · distributed systems · agentic AI
Freelance engineer. Microservices, data pipelines, agentic AI, cloud.
01 / About
I am an engineer with five years of building scalable backends, automation, and AI workflow systems for e-commerce, logistics, and AI products. That work includes 25+ enterprise applications, delivered with teams in the United States and India.
I stay close to the service boundary, the queue, and the deploy. Agents are useful when they call real tools and leave a trail you can debug. The interesting part is the system that keeps working after the demo.
02 / Build your stack
Pick the pieces of the product. I will show the architecture I would actually ship, and the tools I would reach for.
architecture.ts — live
needs
The diagram stays quiet until you pick a need. Then it builds the path I would ship.
what I would build
Choose a need. Services, queues, caches, databases, and agents appear and connect as the scope gets real.
03 / What I can do
Service boundaries, typed APIs, and data models that still make sense after the product changes shape.
Ingest, queues, and workers that move records through the business without someone retyping them.
Orchestrators, MCP tools, and model calls wired into the systems you already run, with a trail you can debug.
AWS infrastructure, deployment automation, and the caching and indexing that keep a hot path honest.
Catalogs, payments, notifications, and the third-party APIs a marketplace actually depends on.
04 / Selected work
A few systems with the outcome written in plain text.
Reverse logistics · 2022 — 2023
Returns across a large retailer network were manual, slow, and easy to abandon before pickup.
Pickup completion up 200% across 2,000+ retailers.
Criminal background verification · 2021 — 2022
A legal-records corpus this large cannot be searched by scanning the primary database on the request path.
700M records, sub-100 ms queries, 10K+ daily users.
2020 — 2021
A marketplace for high-value watches needed payments, notifications, and reporting without a pile of manual admin work.
Manual administrative effort down 80%.
Open slot · Available · open slot
Multi-step agent work needs tools, memory, and a human handoff. Most teams have a demo, not a system.
Open for a first design and a production pilot.
05 / Experience
Jul 2023 — Present
India
Apr 2022 — Jul 2023
Remote, USA
Jun 2021 — Jan 2022
India
Jul 2020 — Apr 2021
India
06 / Skills
AI Powered Glasses for Visually Impaired Person
Published research paper, International Journal of Engineering, 2020
Problem Solving (Intermediate)
HackerRank
AWS Fundamentals: Going Cloud-Native
Coursera
07 / Hire me
Open to freelance engagements and full-time roles. Email works either way.
sandip@build:~
sandip@build — type help
Open to freelance engagements and full-time roles.