Distributed systems
An event-driven dispatch system that matches riders to nearby drivers in real time, with Kafka between every stage and Redis GEO for matching.
10,000+ ride events a minute, P90 driver assignment under 45 ms.
Friedrich, 1818About
- I’m a final-year Computer Science student at NIT Raipur, and I learn best by building a system first and taking it apart second.
- Right now I’m a software engineering intern on Culinda’s AI platform, where I took a slow answer path from 2.8 s down to 120 ms.
- Most of what I build lives behind an API: distributed systems, retrieval pipelines and agents that need guardrails before anyone should trust them.
- Competitive programming is how I unwind. I’m top 2% on LeetCode and a Specialist on Codeforces.
- I’m open to internships and full-time roles, and always up for collaborating on something worth building.
- 2108LeetCode rating, top 2%(LeetCode profile, opens in a new tab)
- 1483Codeforces max, Specialist
- 86thof 27,000+, LeetCode Biweekly 169
- 2ndAI track, Hack-a-Sol hackathon
GitHub activity
@coderconnoisseur(opens in a new tab)1,095 contributions in the last year. Longest streak 15 days.
Experience
Where I’ve worked and studied.
Culinda
Apr 2026 to present
Software Engineering Intern, AI Platform
Remote
- Ship Python and FastAPI microservices for Culinda’s ITSM and contract lifecycle products, scaling async services to 4× throughput.
- Built a Redis semantic cache over vector embeddings that cut P95 latency from 2.8 s to 120 ms, and automated indexing of 10,000+ documents.
- Lowered LLM inference cost by 35% and lifted RAG answer relevance by 18%, checked against an offline evaluation harness.
National Institute of Technology, Raipur
2023 to 2027
B.Tech, Computer Science and Engineering
Raipur, India
- CGPA 8.06 / 10.
- Documentation Executive at the Association of Computer Engineers (ACE), the department’s technical club: I maintained its technical documentation and wrote up events and workshops so the next batch could pick up where we left off.
Projects
A few things I’ve built, most of them to find out how something really works.
AI product
An AI finance platform that reads invoices from photos and PDFs, flags anomalies, and routes plain-English questions to a SQL agent, RAG or an analytics orchestrator.
Runner-up, AI track at Hack-a-Sol, IIIT Naya Raipur.
LLM safety
One LLM agent wrapped in four defense modules, each adapted from an agent-safety paper, and scored before and after with an index I designed.
Prompt-injection success 0.07 → 0, harm score 0.25 → 0, 446 offline tests.
include/spatio_index_core.hpp
Systems in C++
A spatio-temporal index for “where and when” queries: a KD-tree for space, a multimap for time, and Python bindings through pybind11.
Radius and bounding-box queries with optional time windows.
Real-time audio
An always-on-top overlay that captions anything your computer plays, fully offline with Vosk, plus a benchmarking suite for accuracy and latency.
Live captions with no cloud round trip.
internal/cache/cache.go
Distributed systems
A highly available distributed cache written from scratch in Go, starting with a sharded, lock-striped in-memory store.
256 shards, each with its own RWMutex, picked by an FNV-1a hash.
Stack
Tools that have earned a place in the toolbox.
- Python
- Go
- C++
- SQL
- FastAPI
- Flask
- Gin
- SQLAlchemy
- REST APIs
- Microservices
- PostgreSQL
- Redis
- SQLite
- ChromaDB
- Kafka
- Docker
- Prometheus
- Linux
- Git
- CI/CD
- Postman
- Pytest
- RAG
- Vector search
- Multi-agent systems
- LangChain
- OpenRouter
You made it to the end.
Send me your slowest endpoint.
An internship, a full-time role from 2027, or a side project worth building together. My inbox is open.