Mar 2026 – Present · 2 mos · Bengaluru · Hybrid

• Multi-Agent AI Platform for Construction Intelligence
• Architected and deployed a production-grade multi-agent RAG system enabling natural language access to construction project data.
• Built modular orchestration layer for multi-step agentic workflows.
• Implemented dynamic code generation with secure execution in a custom Python sandbox for structured data analysis.
• Designed offline evaluation framework measuring retrieval relevance and response quality to guide prompt and retriever optimization.
• Optimized P95 latency using semantic caching and asynchronous retrieval strategies.
• Implemented structured logging, traceability, and failure monitoring for agent and execution debugging.
• Deployed scalable solution on Azure using Docker and CI/CD pipelines, enabling reliable low-latency production inference
• Impact: Reduced manual data exploration time by 80%, accelerating executive reporting and project decision cycles

Mar 2025 – Mar 2026 · 1 yr · Bengaluru · Hybrid

• Time-Series Forecasting for Workforce Allocation
• Designed forecasting pipeline on ~4 years of utilization data (~1M records).
• Engineered temporal, seasonal, and lag-based features to model project demand patterns.
• Evaluated ARIMA, XGBoost, and LSTM; selected XGBoost for lower variance and stronger cross-validation stability.
• Reduced MAPE from 18% baseline to 7.9%.
• Built rolling-window validation and monthly retraining pipeline.
• Partnered with project managers to operationalize forecasts into staffing dashboards.
• Impact: Improved workforce utilization efficiency by ~12% across active projects

Aug 2023 – Mar 2025 · 1 yr 7 mos · Bengaluru · Hybrid

• Real-Time Handwritten Claims Processing System
• Co-designed and deployed a real-time document intelligence system handling 10,000+ documents/month.
• Built end-to-end workflow using Python, Azure Machine Learning, and Azure AI Services for document ingestion and text extraction.
• Developed rule-based validation engine to handle edge cases, enforce business constraints, and reduce downstream manual review.
• Implemented confidence scoring, monitoring, and feedback loops for continuous model improvement
• Optimized pipeline for low-latency processing, reducing turnaround time from hours to under one minute
• Impact: Achieved 98% precision, significantly reducing manual review workload and operational cost.

Jan 2023 – May 2023 · 4 mos · Karnataka, India · On-site

• Automated Face and Number Plate Recognition and Masking
• Developed and deployed a scalable AI solution using Python, Azure ML Studio, and Azure Computer Vision to detect and mask faces and number plates in images, scaling for 300K+ images with 97% accuracy, and built a monitoring and observability platform to ensure performance tracking and production reliability.