Software engineer at Juspay, where I build real-time data infrastructure in Rust: a Kafka-driven ingestion pipeline that streams payment events into Cassandra, ClickHouse, PostgreSQL and S3 for 50+ services across 14 partner banks. Before that I interned at Amazon, shipping an outage dashboard that went live on Prime Day across 12 marketplaces and is now used by 250+ engineers.
I care about systems that stay correct under load: idempotent writes, dead-letter queues, zero-downtime migrations, and the boring observability that makes 3 a.m. pages short. Lately that interest has turned into building AI agents that can take the 3 a.m. page instead, with enough guardrails that I'd trust one near production.
B.Tech (IT) + MBA, ABV-IIITM Gwalior · Mumbai, India
Languages Rust Java Python TypeScript SQL Bash
Data & streaming Kafka ClickHouse Cassandra PostgreSQL Redis MongoDB DynamoDB
Infra Kubernetes Helm Docker AWS GCP Prometheus Jenkins GitHub Actions
AI LLM agents tool calling RAG LangChain Pinecone Groq
| Project | What it is | Stack |
|---|---|---|
| AI On-Call Engineer | An LLM agent that triages Prometheus alerts, investigates a Kubernetes cluster through 6 read-only tools, and then fixes, proposes a patch, or escalates. 13 write actions sit behind 4 permission tiers and 8 safety guards, enforced in both code and RBAC. Ships with an evaluation harness that injects 50 reproducible faults into a throwaway cluster and scores from cluster state and audit logs, never from what the agent claims. | Python, Groq, Kubernetes, Helm, Prometheus |
| CacheFlow | A multithreaded distributed cache in Java using consistent hashing, request collapsing and prefetching, with pluggable eviction and write policies. | Java, Redis, MySQL |
| AeroSense | Time-series forecasting of air quality with interactive dashboards. | Python |
-
At Juspay, A Rust ingestion pipeline handling 15+ payload types, with batched multi-store writes that cut end-to-end latency by 90% and ClickHouse query time by 60% at 600 rps peak.
-
An analytics platform on Kubernetes CronJobs processing 40M+ events a day into 7 fact tables and partner-facing dashboards.
-
A zero-downtime schema migration on live payment data, using dual writes, reconciliation and safe rollback.
-
Salted SHA-256 protection for sensitive customer identifiers, with zero-downtime key rotation across all services.
-
40+ integration tests and CI pipelines on Azure DevOps and Jenkins, plus UAT with partners before go-live.
-
At Amazon, an outage dashboard built with React, Redux and Node.js, adopted by 250+ engineers and launched on Prime Day across 12 marketplaces for real-time outage response.
-
Consolidated 300+ mitigation levers and outage scenarios into a single MongoDB-backed source of truth, improving on-call efficiency by 40%.
-
An intelligent search with auto-suggestions and advanced filters over that data, supporting 50,000+ concurrent queries.
LinkedIn · LeetCode · nirupamasingh8449@gmail.com
Most of my day-to-day code lives in private repositories at work, so the activity graph here understates things. The projects above are the ones I'd be happy to walk through in detail.