nithin sengodan

Led a structured QA pipeline for AI agentic workflows, adding multi-stage validation and controlled production releases. Extracted monolithic workflow logic into microservices and deployed serverless multimodal inference on AWS Lambda, including a Qdrant-backed semantic search using 1024-dimensional vectors and streaming progress updates.

Built a real-time AI Agent Assist platform integrating multiple data streams and LLM-driven insights for live customer support. Developed distributed transcription and speech-to-speech services with WebSocket and gRPC, optimized Redis pipelines to reduce network latency by 25%, and implemented an air-gapped ASR pipeline processing 45–90 hours of audio per hour of compute while mentoring 5–10 engin

Led end-to-end development of a keyword detection pipeline for air-gapped systems using a scalable microservices-based framework. Improved internal TTS inference performance by 20% via ONNX conversion and runtime optimization, built a multilingual video translation pipeline for NPTEL across 11 languages, and developed secure biometric dashboard backends with encryption and RBAC.

Designed real-time water consumption analytics systems for the SIPCOT Smart Initiative, building time-series pipelines and monitoring dashboards with InfluxDB and Grafana. Implemented an OCR-based medical report digitization platform achieving over 90% accuracy and processing reports in under 6 seconds on a single-core machine.

Conducted data scraping and cleaning for marketing datasets, improving lead quality and automating validation pipelines to streamline downstream workflows.

Completed advanced machine learning coursework with Stanford University via Coursera, covering topics such as anti-spam filtering, image recognition, clustering, and recommendation systems.