GUSAIN
Backend-focused engineer who likes building systems that handle real-world scale and make people's lives easier. Currently at Neurofin AI building LLM inference services and high-throughput data pipelines. Core strengths: Python, Go, distributed systems, databases, and LLM engineering.
Off the clock: football, hiking, and reading — lately Russian literature (Chekhov, Dostoevsky).
- Shipped LLM-powered document boundary detection (90%+ accuracy) and classification services in production — event-triggered via AWS SQS/SNS, orchestrated as Temporal activities (retries, deterministic recovery, stage-level observability), routed across Gemini 2.5 Flash & Pro on Vertex AI via LiteLLM, traced with Langfuse; cut per-call token usage ~40% via prompt caching + context engineering.
- Built an ETL pipeline migrating 5M+ documents/day from MongoDB to PostgreSQL — powers Apache Superset dashboards used daily by 7 lending-ops teams; internal AI chatbot queries the warehouse via a custom MCP server.
- Delivered 20+ full-stack features end-to-end (React + Python/FastAPI) using Claude Code.
- Cut critical API p99 latency 80% (2.5s → 0.5s) via Redis caching (TTL/LRU), SQL CTEs, MongoDB aggregation pipelines, Flask tuning.
- Built Kafka event streams + Celery/Redis async pipelines (monitored via Flower); batch-processed millions of rows (30s → 3s); shipped an FCM notification system for 1000+ users.
- Delivered real-time partner dashboards for Hyundai, Kia, and BMW handling 40,000+ tickets; shipped Chargemate (FastAPI) for 100+ field technicians.
- Built Cristalina — internal ticketing/ops platform (Flask) for 1000+ users across Ops, Finance, SCM, Support; retired $12K/yr in SaaS spend.
- Reduced AWS costs 25% via lifecycle rules, right-sizing, automated cleanups.
- Optimized SQL for 10+ infrastructure tables.
- Automated Excel reporting for 3 departments.
Open-source, self-hostable notification infrastructure. One trigger API fans out to 8 channels. A drag-and-drop visual workflow builder changes routing without redeploying.
- In-App
- Slack
- Discord
- Telegram
- SMS
- Push
- Per-channel Celery queues; real-time in-app via WebSockets + Redis Pub/Sub
- Fernet-encrypted per-team credentials (BYOC — bring your own credentials) & scoped API keys
- Ships Node & Python SDKs in-repo; fully containerized (docker compose up)
Event-driven backend demoing the producer/consumer pattern. A token-authenticated API submits vendor-order events → queued through Redis Streams → an async FastAPI consumer persists to SQLite → aggregates per-vendor metrics and can render a revenue plot. Swagger docs at /docs.
A URL shortener built as a system-design exercise. Base62 short codes (encode the DB row id); PostgreSQL primary store; Redis LRU cache for hot links. README includes capacity estimates (100M links ≈ 10 GB).
Lightweight macOS menu-bar clipboard manager. Auto-capture, dedup, and debounced search; ships a .dmg.