GGSIPU, Delhi · 89%
Vasu Sharma
Senior Software Engineer
From distributed backends and Kafka pipelines to RAG systems running in production — I build platforms for scale, down to the infra and compilers they stand on.
- Backend & Distributed Systems
- AI / ML Platforms
- Compilers & Infra
Focus
Five years shipping production systems end to end. My work moves between systems beneath the surface and platforms built for scale — and I like keeping a foot in each.
Compilers & runtimes
Deep in LLVM: porting MemorySanitizer to AIX/PowerPC64, designing shadow-memory structures and a custom heap allocator, writing platform interceptors, and contributing upstream. Close to the metal, where correctness is not optional.
Backend & AI platforms
API integration platforms, event pipelines on Kafka and Redis handling a million-plus concurrent requests, double-entry accounting ledgers, and RAG systems that classify failures in production — built to hold up under real load.
Experience
Senior Software Engineer
- Built an AI regression-triage system — a RAG pipeline that ingests unstructured test logs, embeds them into a vector store, and auto-classifies failures by root cause, cutting manual triage time by ~40% across thousands of cases.
- Shipped BuildMonitor, an AI merge-conflict resolver: led discovery, built the proof-of-concept, drove org-wide adoption, and mentored integrators — cutting branch-reconciliation time by ~20%.
- Ported MemorySanitizer to AIX: designed shadow-memory data structures, a custom heap allocator, and platform-specific interceptors, scoped into phased deliverables under tight timelines.
- Contributing to LLVM upstream, collaborating asynchronously with distributed maintainers across review cycles.
— Aug 2026
Software Engineer II
- Built an open-source managed service for API integrations in Python and Go, cutting third-party integration time by 60% with 99.9% uptime on Kubernetes.
- Designed API infrastructure with RS256 JWT auth, OAuth 2.0 flows, and RBAC, holding ~50ms latency under audit and data-governance requirements.
- Built Kafka-based async pipelines with Redis caching handling 1M+ concurrent requests, at 90% test coverage with pytest.
— Nov 2024
Software Engineer I → II
- Built the Class & Contest Creation platform (monolith backend + micro-frontends), dropping error rates from 7.5% to 0.1% and saving 40% of content-ops hours.
- Shipped Course Pause/Resume end to end with a double-entry accounting ledger and idempotent APIs — ~50% fewer support tickets and +15% revenue retention.
- Led RCAs on production breakages, optimized hot-path queries, and set CI/CD standards with Prometheus monitoring across services.
— Jun 2024
Selected work
AI Job Search Agent
An autonomous agent that searches openings, scores role fit, tailors the résumé for each, and fires alerts on strong matches.
Coverage Gap Scanner
Finds untested paths in cross-platform C/C++ codebases through static analysis, then uses LLMs to triage what the gaps actually mean.
Deterministic Motion Renderer
A render pipeline that guarantees byte-identical motion-graphics output for the same input, exposed behind a REST API.
Distributed Rate Limiter
A sliding-window rate limiter backed by Redis, built to stay correct and fair under heavy concurrency.
Toolkit
Writing
Notes on compilers, backend systems, and the occasional detour into philosophy and shayari.
Read on MediumGet in touch
Off the clock, I read Camus and the Stoics and write the odd bit of shayari. I like problems that sit close to the metal and refuse to stay abstract — and I care about work I can stand behind and explain.
GGSIPU, Delhi · 89%