Bengaluru, Karnataka, India
AI/ML professional with nearly 15 years of experience in software engineering, including over 9 years of expertise in applied machine learning, deep learning, and generative AI. Holds an M.Tech in Artificial Intelligence and Machine Learning from BITS Pilani. Currently serving as AI & Data Technical Lead at the Innovation Hub, Bangalore, driving AI-powered innovation and data-driven solutions. My expertise lies in designing and deploying production-grade ML systems at enterprise scale — spanning transformer-based document intelligence, RAG-powered conversational AI, multi-horizon time-series forecasting, and retail ML. I operate across the full ML lifecycle: from research and experimentation through to CI/CD-enabled production deployment and ongoing MLOps. Technically, I work extensively with Python, PyTorch, Hugging Face Transformers, LangChain, MLflow, and a range of LLMs including GPT-4, Claude (Anthropic), LLaMA, and Mistral. I have deep hands-on experience with Azure ML, AWS SageMaker, AKS, Docker, and Kubernetes — and I am comfortable building both real-time inference pipelines and large-scale batch processing systems. Beyond individual delivery, I lead cross-functional teams across geographies, aligning technical strategy with product and business objectives. I take pride in translating complex AI research into solutions that generate measurable business value — whether that is a 60% reduction in document onboarding time, a 15–20% improvement in forecasting accuracy, or a 94% precision fraud detection system. I am always open to conversations around AI/ML innovation, enterprise AI architecture, and opportunities to collaborate on challenging problems.
Leading the architecture and end-to-end delivery of enterprise-grade AI/ML systems across document intelligence, conversational AI, and predictive logistics analytics. Responsible for the full ML lifecycle — from ideation and experimentation to production deployment and ongoing model governance. Key projects: ▸ Document Intelligence Platform: Architected a production-scale document extraction system using LayoutLMv3 and transformer-based vision-language models to process complex unstructured documents (Bills of Lading, invoices, shipping documents). Built dual-mode OCR and OCR-free inference pipelines on Azure ML + AKS, with MLflow-based experiment tracking, automated retraining, and A/B testing frameworks. Reduced new document type onboarding time by 60%. ▸ Trip Connect – AI Travel Assistant: Engineered a RAG-powered multi-modal travel assistant combining BM25 lexical ranking, knowledge graph traversal, and semantic search (Qdrant + sentence-transformers) via LangChain Agents and ReAct patterns. Delivered sub-second response latencies with high answer relevance, and designed intelligent escalation logic to reduce Tier-1 support ticket volume. ▸ Container Demand Forecasting: Designed and trained Temporal Fusion Transformer (TFT) models for multi-horizon logistics demand forecasting. Engineered comprehensive time-series feature suites and achieved a 15–20% RMSE reduction over prior models, directly improving container utilisation and capacity planning. Stack: LayoutLMv3 · PyTorch · LangChain · RAG · Qdrant · TFT · Azure ML · AKS · MLflow · Docker · Hugging Face · Python
Transitioned from L3 development lead for enterprise retail POS platforms into an applied ML engineering role, embedding machine learning capabilities into retail store systems for global clients including Sephora, Singapore DFS, and CPW. ▸ Purchase Propensity & Personalisation Engine (Sephora · Singapore DFS): Built and deployed a real-time XGBoost/LightGBM propensity scoring model integrated into the TP.net POS platform, surfacing personalised product recommendations at checkout. Increased average basket size by ~18%. Engineered customer segmentation pipelines using K-Means clustering on transactional data. ▸ Demand Forecasting & Inventory Optimisation (CPW): Built multivariate ARIMA/SARIMA + Gradient Boosting ensemble models for SKU-level demand forecasting. Reduced overstock incidents by 22% and out-of-stock occurrences by 15% within six months of deployment. ▸ POS Fraud Detection (Singapore DFS): Designed a supervised fraud detection system using Random Forest and Gradient Boosting classifiers achieving 94% precision at under 2% false positive rate. Engineered 40+ discriminative features from POS transaction logs and delivered a Power BI fraud analytics dashboard for loss prevention teams. Stack: XGBoost · LightGBM · Random Forest · ARIMA/SARIMA · Scikit-learn · SSIS · SQL · C# · .NET · Power BI · Python
Delivered analytics engineering and application development for enterprise financial data products in the investment management industry. Built data pipelines and reporting solutions for MSCI's Lease Intelligence platform, improving data throughput and report generation speed by 35%. Developed automated server health-check utilities and secure data export modules, maintaining 99.9% uptime SLAs for critical financial services. Stack: ASP.NET · C# · SQL Server · SSRS · ADO.NET · WPF · VBA
Delivered bespoke enterprise web applications for global healthcare clients, primarily Baxter Healthcare UK. Built data-driven modules for supply chain management and inventory tracking, and developed integration components with AS400 legacy systems. Collaborated with cross-functional teams to ensure on-time delivery against contractual SLAs. Stack: VB.NET · Windows Forms · MSSQL Server · AS400