Pune Division, Maharashtra, India
I build production-grade AI systems that move beyond experimentation to deliver measurable business outcomes — across machine learning, optimization, and generative AI. Over the past 5+ years, I’ve worked on end-to-end AI systems in industrial, supply chain, and pharma environments, where success depends not just on model performance, but on scalability, reliability, and real-world decision impact. My work spans: Machine Learning: predictive modeling, forecasting, NLP, and optimization (XGBoost, CatBoost, regression, time series) Generative AI: LLMs, RAG pipelines, agentic workflows, and enterprise knowledge systems (LangChain, LangGraph, OpenAI/Azure OpenAI) Production Systems: MLOps (MLflow, CI/CD), FastAPI services, Docker-based deployments, scalable cloud architectures (Azure) I focus on building systems that integrate into business workflows, not just models in isolation — translating ambiguous problems into deployable AI solutions. 💡 Selected Impact: • Delivered $850K+ cost savings through supply chain optimization models • Reduced downtime by 20–25% using predictive maintenance systems • Enabled $100K–$300K/month revenue recovery via LLM-powered automation • Built AI systems serving 4,000+ daily users across 20K+ sites ⚙️ Core Focus Areas: • Production ML systems & scalable AI architectures • Generative AI (LLMs, RAG, multi-agent systems) • Decision intelligence & optimization • End-to-end system design (problem framing → deployment → monitoring) 🏗️ Experience: Currently building ML and agentic AI systems at Johnson Controls; previously worked at General Mills, TIBCO, and Wynum across large-scale enterprise use cases. 🎓 Background: M.Tech in Modelling & Simulation (Machine Learning, Operations Research, Statistics)
• Built a production-grade conversational AI chatbot using LangGraph multi-agent architecture, combining NLP-to-SQL and RAG pipelines to deliver real-time business insights — serving ~4,000 daily users across 20–30 branches • Built an end-to-end service request automation system using LLMs, fuzzy matching, and LLM-based reranking — automatically parsing customer emails, extracting site, priority, and issue details, creating SRs, and dispatching confirmation back to customers; reduced processing time from 5–10 minutes to under 1 minute across 1,000+ daily emails using Azure Functions and Power Automate • Built a revenue recovery pipeline using LLMs to identify unbilled service requests, generating $100K+/month in additional revenue with zero manual intervention • Deployed predictive maintenance models (CatBoost + SHAP) to detect equipment faults early, reducing operational downtime by 20% across field service operations • Owned end-to-end delivery of 3–4 production AI services — from model development to deployment via FastAPI, Docker, Azure App Services, and Azure DevOps CI/CD pipelines
- Built analytical solutions using Python and BigQuery to identify supply chain inefficiencies and unmet demand - Analyzed labor impact on machine performance using regression models to optimize operational efficiency - Developed optimization models for production planning, resulting in $850K+ cost savings - Collaborated with cross-functional stakeholders to drive data-driven decision making
- Built predictive models to classify semiconductor wafer defects using SVD, clustering, and signal processing techniques - Applied machine learning to identify root causes of manufacturing defects and improve yield - Delivered data-driven insights for process optimization in semiconductor manufacturing
- Built ETL pipelines and NLP-based systems to extract insights from clinical trial data - Developed predictive models using regression techniques to reduce maintenance downtime - Applied advanced techniques such as Fourier Neural Operators for time-series prediction
Title 'Design and Development Of Pendulum Bumper Impact Test Rig'