GC OpenSoft’25 (Gold) – Deloitte Sponsored Hackathon | “Counselor Chatbot for Employees” IIT KHARAGPUR Mar 2025 • Designed a three-stage agentic AI pipeline integrating time-adjusted EMA, Isolation Forest, and anomaly detection models to flag employee well-being risks from behavioral and activity data.• Engineered modular microservices in FastAPI and LangGraph for data ingestion, LLM-driven reasoning, and HR analytics, ensuring real-time interaction between chatbot and analytics engines.• Implemented parallelized report generation using Agno and LanceDB, enabling concurrent multi-agent… Show more • Designed a three-stage agentic AI pipeline integrating time-adjusted EMA, Isolation Forest, and anomaly detection models to flag employee well-being risks from behavioral and activity data.• Engineered modular microservices in FastAPI and LangGraph for data ingestion, LLM-driven reasoning, and HR analytics, ensuring real-time interaction between chatbot and analytics engines.• Implemented parallelized report generation using Agno and LanceDB, enabling concurrent multi-agent processing and reducing report generation time by >40%.• Developed an adaptive conversational chatbot capable of session management, memory retention, and context tracking, deployed as a scalable containerized service via Docker and Kubernetes.• Integrated custom LLM pipelines for context-aware dialogue, automated anomaly explanations, and mood-based insights; optimized API load balancing to handle 100+ concurrent sessions.• Achieved a 50% reduction in manual HR analysis effort, delivering an AI-powered counselor that autonomously identifies employee concerns and generates actionable summaries for HR teams. Show less

Among Top 52 Teams from 1000+ teams across India| Round 2 Convolve 3.0 IIT GUWAHATI Feb 2025 We created an AI solution that developed a Behavior Score Model for Bank A's Credit Card customers, predicting the probability of future defaults. This risk management framework enabled the bank to proactively mitigate credit risks across its portfolio. Using a dataset of 96,806 development samples, we trained a predictive model to generate default probabilities, which were then applied to 41,792 validation samples for risk assessment. Our team was selected for Round 2 and ranked among… Show more We created an AI solution that developed a Behavior Score Model for Bank A's Credit Card customers, predicting the probability of future defaults. This risk management framework enabled the bank to proactively mitigate credit risks across its portfolio. Using a dataset of 96,806 development samples, we trained a predictive model to generate default probabilities, which were then applied to 41,792 validation samples for risk assessment. Our team was selected for Round 2 and ranked among the top 52 teams out of 1000+ teams across India.We did not get selected for the final round but it was a learning experience throughout. Show less