Backend engineer with 4+ years building and scaling production systems β a 0β1 fraud-detection network over 10M+ data points, an e-commerce platform serving 1M+ users, and a zero-downtime MongoDB β ScyllaDB migration that cut ~$150K/yr. Codeforces Candidate Master. These days a lot of my time goes into finding and fixing subtle correctness bugs in large open-source AI systems β agent frameworks, inference engines, eval harnesses, and vector/data infrastructure.
π SDE2 @ Uni Cards (fintech) Β Β·Β π Bengaluru, India Β Β·Β π¬ Ask me about backend architecture, distributed systems & databases
Self-found, regression-tested fixes across the AI stack β every PR ships with a fail-before / pass-after test:
| Project | Stars | Contribution | Status |
|---|---|---|---|
| KubeRay Β· CNCF / Ray | 2.6k β
|
Gang-scheduling resource accounting for suspended worker groups β Go | β Merged |
| Keep Β· YC W23 | 12k β
|
Percentage alert thresholds were compared as strings β missed alerts β Python | β Merged |
| Ollama | 175k β
|
Tool-call parser silently dropped calls with braces inside string args β Go | π΅ Open |
| mem0 Β· YC S24 | 60k β
|
xAI provider crashed under HTTP-proxy configuration β Python | π΅ Open |
| LlamaIndex | 50k β
|
Vector-store metadata filter crashed on its own default condition β Python | π΅ Open |
| OpenAI Agents SDK | 27k β
|
Per-request token accounting when providers omit total_tokens β Python |
π΅ Open |
| OGX Β· ex-Meta Llama Stack | 8k β
|
Keyset-pagination silently dropped rows in the SQL store β Python | π΅ Open |
| MCP Go SDK Β· Google Γ Anthropic | 5k β
|
Streamable-HTTP transient reconnect poisoned the session β Go | π΅ Open |
| Mercor Β· APEX-Agents | 201 β
|
Grading-integrity fix in the agent evaluation harness β Python | π΅ Open |
- Languages β Java Β· Python Β· C++ Β· SQL Β· Go
- Backend & Distributed Systems β Spring Boot Β· Microservices Β· Temporal Β· Kubernetes Β· Docker Β· AWS Β· Kafka Β· PostgreSQL Β· MongoDB Β· ScyllaDB Β· DynamoDB Β· Redis
- AI / LLM β LLM agents & tool-calling Β· MCP Β· inference & serving internals Β· eval / grading harnesses
- Codeforces β Candidate Master Β· Adi_7861
- CodeChef β 5β Β· adi_7861 β Global Rank 45, Div-1 Long Challenge
- TCS CodeVita β Global Rank 283 / 10,000+
SDE2 Β· Uni Cards (fintech) β fraud-detection systems, e-commerce platform (1M+ users), MongoDB β ScyllaDB migration. Previously Vahan (YC S19) and Optmyzr.


