forofficialsanjit@gmail.com
Sanjit Shrestha!Feel free to explore my work and reach out — I architect systems that pay people, move freight, and stay up.
We start with a call to map your goals, constraints, and the failure modes that scare you most. I leave with a clear picture of what "done" actually means.
Architected the payments platform for a US-based consumer lender. Designed the ledger, idempotency keys, ACH/Plaid/Stripe rails, and the reconciliation pipeline. Led a team of 6 backend engineers from kickoff to production.
Real-time tracking and automated invoicing for a regional freight carrier. Decomposed a four-year-old monolith into 14 services around shipments, billing, and notifications. 1M+ events/day across the bus.
Analytics surface that turns raw event streams into plain-English summaries and anomaly alerts. Built the prompt-cache layer, the cost governor, and the JSON-mode validator. Cut model spend 40% in the first quarter.
Bespoke gateway in front of 22 backend services. Token-bucket rate limiting, circuit breakers, OpenTelemetry across the board. Handled 1M+ requests/day at single-digit ms p99 overhead.
Led the breakup of a Django monolith into Spring services on AWS. Strangler-fig pattern across 9 months. Zero-downtime cutover, regression-test suite green throughout, no rollbacks.
Architecture ownership for Big4-aligned engagements. I lead cross-functional delivery teams, define engineering standards, and drive the AI-integration roadmap across the org. Hands-on with the systems I design — code, reviews, and on-call rotations.
Led the engineering team. Designed microservices handling millions of requests per day. Established CI/CD, observability, and code-review practices that survived my tenure.
Built backend services and APIs serving thousands of daily users. Spring Boot, PostgreSQL, REST. Where I learned that a clean migration history is worth more than any clever framework.
A written audit of your current system: bottlenecks, failure modes, and the changes that buy you the most runway.
Strangler-fig plan and execution. The cutover playbook I've used to break apart four monoliths without a rollback.
A prompt-cache, budget guard, and JSON-mode validator dropped in front of your LLM calls. For teams whose model bill outpaces revenue.
If you're building something that needs to stay up — payments, logistics, AI on real data, or an architecture that's stopped scaling — I'd like to hear about it.