Building autonomous intelligence systems — from Computer Vision pipelines to LLM-powered agents that operate, reason, and act in the real world.
I'm Mayur Shirsat — an AI/ML Engineer, based in Pune, Maharashtra. I specialise in building production-grade AI systems that bridge research and reality. From fine-tuning large language models to deploying computer vision pipelines on edge hardware, I engineer solutions that don't just prototype well — they scale.
With a Master's in Data Science from Symbiosis International University and a foundation in Computer Science from Bharati Vidyapeeth, I bring rigorous academic thinking to hands-on engineering. My work spans agriculture, real-estate, healthcare, and education — always chasing the frontier of what autonomous intelligence can do.
Five production-grade AI systems — each solving a real problem at scale.
Automated logistics ecosystem utilizing YOLOv11/v26 and CNN-LSTM frameworks for 360° damage assessment and machinery tracking. Achieved 99.9% logging accuracy by integrating VLMs for OCR and real-time Digital Twin mapping to eliminate manual yard searches.
LLM/VLM microservices — Gemma-3-4B generates exam papers, Qwen-2.5-VL extracts student answers from scanned sheets, GPT-OSS 20B auto-evaluates and scores through coordinated microservice architecture.
View on GitHub →YOLO-based real-estate damage detection — wall cracks, fire damage, rusting, wire failures — with 90%+ mAP. Defect alerts trigger VLM verification and automated draft emails to builders.
Cross-modal advisory platform synthesising text, image, and speech inputs via LLM-driven reasoning. Achieved 95% accuracy in plant disease identification. Integrated Hugging Face TTS + FastAPI for real-time market data and 5 government welfare scheme access.
Power BI dashboard backed by SQL — tracks KPIs across departments, roles, and audiences. Corporate projects identified as top revenue contributor at $510.75M. Open-sourced on GitHub.
RAG system using Meta LLaMA — extracts and summarises data from 5+ document types (PDF, DOC, CSV, links, text). Integrated NetworkX Knowledge Graph RAG. 90% accuracy, 40% better document analysis.
Have an AI project in mind? Looking for an ML engineer who ships to production? Let's talk.