experience

4 years building production systems across ai, backend, and infrastructure.

Senior Software Engineer September 2024 – Present
Innoviti Technologies · Bengaluru
  • Designed Sanchez — an internal production control platform on Django housing 8 isolated apps with Google OAuth, RBAC, and per-app data isolation; built an LLM-powered log monitoring pipeline with Slack alerting, per-app logical behaviour analytics, and a unified control plane — consolidating all scattered production tooling into a single governed environment.
  • Built the AI inference stack from scratch — vLLM on 96 GB VRAM with tensor parallelism, 8-bit AWQ quantization, and continuous batching behind an async FastAPI gateway; achieved 128K context window locally, 50+ concurrent requests at 3,000+ tokens/sec, sub-3s p99 latency.
  • Built Hermione — a natural language analytics engine on Django + Qwen3-VL 27B (8-bit) on AWS EC2; schema-aware SQL generation with an auto-visualization layer producing charts, graphs, and maps from PostgreSQL — handling 1,600+ queries/day at 92% accuracy.
  • Built Holmes AI — an async image QC pipeline on FastAPI + RabbitMQ processing 5,000+ support ticket images daily; S3 artifact pull, preprocessing and compression, YOLOv9 checkbox detection, Qwen3-VL 32B semantic validation — reducing processing time from 1,260s to 7s (180×).
  • Built Link — a fraud detection system for a 21,000+ merchant network; RabbitMQ event ingestion, feature engineering on transaction history and velocity, LightGBM classifier with tuned decision boundaries, low-latency scoring API at 100,000+ transactions/day — reducing LEA complaints by 78%.
  • Redesigned field engineer dispatch across 20,000+ stores and 120 engineers — weighted bipartite matching with KPI-based priority scoring (SLA, skill, zone), Haversine geo-proximity via OpenStreetMap, priority queues on PostgreSQL with spatial indexing — reducing backlog by 37%.
FastAPI Python vLLM LightGBM AWS PostgreSQL Celery YOLOv9 Django Sanchez
Software Development Engineer July 2022 – September 2024
Pheonix Solutions · Bengaluru
  • Built Disha-AI — a two-stage address normalization pipeline on FastAPI; Google Maps API for geocoding, Gemini 1.5 Flash for semantic correction, Redis-backed deduplication and rate limiting — 94%+ accuracy across 20,000+ records, $28,000 cost reduction.
  • Built a high-throughput inventory system on Django REST + PostgreSQL — optimistic locking for concurrent mutations, write-through Redis caching with TTL invalidation, connection pooling — sustained 12K req/s across 50K+ stock units.
  • Built a compliance portal on Django + Celery — EasyOCR license extraction, pHash duplicate detection, automated MVR checks with audit trails — reduced review time by 30%.
  • Built a social profile intelligence crawler — distributed Selenium with rate-limit-aware scheduling and session rotation, Mistral 7B for structured extraction — 1,500+ profiles/day at 92%+ accuracy with zero manual tagging.
FastAPI Django Redis PostgreSQL Selenium Mistral 7B Gemini
Intern January 2022 – June 2022
KPMG India · Ahmedabad
  • Automated the client license audit pipeline — FastAPI bulk upload layer, PyPDF + EasyOCR extraction, Python normalization into SQLite — cutting processing time by 90% and eliminating manual document review.
  • Unified $7M of fragmented license data from MySQL, Excel, and OCR sources into a single SQLite store via ETL; connected to a Power BI dashboard with utilization gap detection and cost optimization insights for KPMG stakeholders.
Python EasyOCR SQLite Pandas Power BI