Nithin Varghese

I’m an applied ML engineer with 6+ years of experience building and shipping production AI systems end to end. My background is strongest in computer vision and real-world ML deployment, where I’ve led projects from problem framing through rollout, optimization, and operational support across industrial, retail, and banking environments. More recently, I’ve also been building agentic AI systems and production workflows.

I’m most effective in roles where the work goes beyond model-building and into delivering reliable systems that create measurable value for users.

AI Consultant at Self-employed (2024-12 – Present)

Self-employed AI consultant specializing in multi-agent systems and voice-first ideation tools

• Built Storybook Multi-Agent System (latentstory.com) — filesystem-centric LangGraph pipeline generating illustrated storybooks → print-ready PDFs; 6 agents, realtime streaming UI, persistent workspaces
• Built & shipped FlowWriter — voice-first socratic ideation tool that converts conversations into structured outputs (blogs, tweets, SCAMPER, etc.) using LLM agents

Head Computer Vision Engineer at DeepSight AI Labs (2021-02 – 2024-05)

Led end-to-end delivery of multi-site computer vision systems across mining, energy, banking, and retail sectors with focus on production pipelines, edge deployment, and real-time analytics

• Led end-to-end delivery of multi-site CV systems across mining, energy, banking, and retail; owned scoping, architecture, model ops, and on-site rollout
• Built production pipelines for critical use cases (no-PPE, fire, intrusion, crowd/vehicle counts, tailgating, idle time, zone breaches) deployed across major Coal India sites
• Rolled out high-impact verticals: ArcelorMittal 'person-under-steel-roll' safety detection; national exam proctoring analytics scaling to 10,000+ cameras per exam; ATM security (helmet/mask, intrusion, tamper/fault)
• Delivered HPCL wagon tracking/counting to cut manual errors via automated CCTV analytics at industrial plants
• Architected and shipped edge-first, real-time analytics for banking/retail; optimized model throughput/latency and fleet reliability at scale
• Deployed facial recognition with vector-DB retrieval end-to-end (training → indexing → site rollout) for high-accuracy identity workflows
• Spearheaded hardware-aware inference optimization (TFLite, TensorRT, OpenVINO) for Raspberry Pi/CPU/GPU, reducing cost while boosting FPS and stability
• Drove continuous SOTA adoption (architectures, compression, scheduling, monitoring) to improve accuracy, robustness, and scalability

Computer Vision Engineer at Deepsight AI labs (2018-09 – 2021-02)

Developed and deployed computer vision models for various detection tasks including camera tampering, facial recognition, and crowd analysis

• Created Novel algorithm for camera Tamper detection
• Trained and deployed models for masked face, helmet detection
• Trained and deployed models for crowd detection use case

Machine Learning Engineer at Freelancer (2018-05 – 2018-08)

Developed computer vision solution for receipt data extraction

• Receipt Information Extractor: Developed a computer vision solution that processes an image of a receipt to automatically detect relevant data fields using YOLOv3-based object detection, then seamlessly passes the extracted data through an OCR pipeline to produce a structured CSV output

Bachelor of Technology in Information Technology – School Of Engineering, CUSAT (2018-05)