Backend & AI systems engineer. I build distributed backends in Go and Python, and LLM systems that I measure instead of just demo. Currently at Culinda Inc. Open to SWE / AI engineering roles.
Portfolio · LinkedIn · Résumé · nishantborkar28@gmail.com
| Project | What it is | Proof |
|---|---|---|
| Guarded Agent Ensemble | LLM agent wrapped in 4 defenses adapted from agent-safety papers, A/B evaluated with a composite index | 733 tests; on AgentDojo held-out banking, attack success 39/72 → 0/72; reports where the defenses hurt a strong model |
| RidePulse | Event-driven ride matching in Go · Kafka · Redis GEO · Prometheus | ~400 RPS intake, successful-match p95 33–84 ms; written-up race-condition and deadline fixes |
| Lumen | AI money advisor: invoice reading, text-to-SQL agent, hybrid RAG | Live demo · retrieval top-5 hit rate 67 → 100 / 104 with BM25 + reranking |
| EchoLine | Offline real-time captions for anything playing on Windows | Caption latency p50 266 ms / p95 438 ms; installer shipped via GitHub Actions releases |
| SpatioX | C++17 spatio-temporal index (KD-tree + time index) with Python bindings | 5–20× faster than a linear scan at 1k–50k records |
Languages: Go · Python · C++ · TypeScript · SQL
Backend: Gin · FastAPI · Flask · Kafka · Redis · PostgreSQL · Docker · Prometheus
AI: LLM agents & evals · LangGraph · RAG (pgvector, BM25, reranking) · speech (Moonshine)
Competitive programming: Codeforces · LeetCode


