Hello, I'm
I design and deploy production-grade AI systems—GenAI routing, vector search, and computer vision— optimized for accuracy, latency, and cost across AWS, GCP, and Azure.
Building AI products that deliver measurable impact
I work at the intersection of research and production, translating modern ML/LLM capabilities into reliable systems. Recent work includes an LLM-driven "Company Brain" that turns enterprise documentation into agent-executable ERP workflows, multi-cloud LLM routing for cost-efficient GenAI (now running 2 enterprise pilots), embedding-based recommendation with fast vector search, and document understanding pipelines that convert unstructured PDFs into structured JSON.
I care deeply about engineering fundamentals—evaluation, ablation studies, monitoring, and scalable deployment. I also enjoy teaching and mentoring, and have trained hundreds of learners through advanced ML/DL and LLM courses.
GenAI, ML/DL, and production deployment
Routing, prompt/context engineering, agents, fine-tuning, evaluation.
Building robust services for LLM + ML workloads.
Feature engineering, model selection, metrics-driven iteration.
Embeddings, detection, and modern vision architectures.
Low-latency similarity search and graph-grounded retrieval.
Multi-cloud orchestration for AI workloads.
Containerized inference and delivery pipelines.
Practical tooling for production and testing.
Where I've delivered products, research, and training
Vizuara Technologies Pvt. Ltd.
G. H. Raisoni College of Engineering, Pune
Guru Gobind Singh College of Engg & Research Center, Nashik
G. H. Raisoni College of Engineering, Nagpur
A few projects with clear outcomes (public links when available)
GenAI request routing that dynamically selects the best model across Vertex AI, Bedrock, and Azure AI Foundry, with a lightweight selector model fine-tuned using Unsloth. Delivered 74% cost reduction and secured 2 enterprise pilots.
Image-based recommendation using BEiT/ViT/CLIP embeddings with FAISS for low-latency similarity search. Achieved Top-3 accuracy of 0.82 and deployed for scalable inference.
Built a chatbot that uses a knowledge graph + GraphRAG with Neo4j and Cypher queries to ground responses, improving answer relevance and traceability over unstructured retrieval alone.
Accounting platform for Indian SMEs with a 3-tier classification engine (pattern → fuzzy → LLM) auto-categorizing 100+ vendors across 21 expense heads, exports to Tally Prime & Zoho Books, and LLM invoice extraction at 94%+ accuracy.
LLM system that extracts business logic from enterprise documentation, validates it against stored procedures, and generates agent-compatible executable journeys for autonomous ERP workflow execution and cross-version data migration.
Built a high-accuracy pipeline to convert unstructured PDFs into structured JSON for API integration. Benchmarked GPT, Gemini, and open-source models; achieved 91% accuracy.
Implemented object detection using YOLOv11 and YOLOv11-OBB. Achieved F1 score of 0.87 and optimized the pipeline for deployment.
Certifications, education, and community impact
Have a GenAI/ML problem to solve? Let's talk.
