Systems that run without handholding.

GRC Virtual Assistant (Compliance Chatbot)

• Built a conversational AI platform for raising compliance and access requests in natural language — 607 unique users, 1,145+ conversations since launch.
• Set up a RAG pipeline over internal compliance docs, with particular care on preserving table structure during ingestion to keep answers accurate on structured data.
• Facilitated 62 GRC access requests end-to-end through chat, replacing manual form-based workflows.
• Integrated the internal LLM (Nexus) with prompt logic for follow-ups, conversation context, and role-based responses.
• Drove 4.7x adoption growth over four months — 77 → 193 → 317 → 362 conversations (Dec–Mar).
• Built the full Flask backend: 20+ REST endpoints, session management, PostgreSQL.
• Added browser-based language detection so the bot replies in the user's preferred language.

Flask · PostgreSQL · RAG · Nexus LLM · Prompt Engineering

Automated Incident Allocation System (Ticket Router)

• Pulls ServiceNow incidents every 5 minutes and assigns them by agent shift, availability, and current load.
• Min-heap-based fair distribution across P1–P5 priority tiers — cut agent workload by ~40% and keeps SLA-critical P1/P2 tickets assigned within seconds.
• Architected as three modular Airflow DAGs: shift roster sync (MinIO), a 5-minute allocation cycle (~288 runs/day), and end-of-day redistribution of unresolved tickets.
• Handled IST/CET timezone and DST logic to support overlapping India–Germany shifts.
• Integrated 4 ServiceNow API endpoints for ingestion, assignment, resolution updates, and audit trails — full traceability for management review.
• Currently building a resolution knowledge base with OpenAI embeddings + RAG to recommend fixes from historical patterns, targeting a 20% reduction in mean time to resolution.

Apache Airflow · ServiceNow API · OpenAI Embeddings · RAG · MinIO

• Took over recurring weekly/monthly reports and automated them with SQL and Python, cutting manual effort by ~35%.
• Analyzed shop-floor data — downtime, utilization, yield — surfacing patterns behind a 12% improvement in equipment efficiency.
• Built a Financial Report Analyzer (Python + OpenAI) that reads PDFs, extracts key metrics, and generates summaries — saved the team ~40% of analysis time.

Python · SQL · OpenAI API · Power BI

• Built NovaFit, a full-stack fitness tracking app (MERN) that grew to 100+ active users.
• Wrote RESTful APIs for authentication, data storage, and retrieval.
• Helped Tamaku NGO digitize their records — retrieval went from slow and error-prone to 50% faster.

MongoDB · Express · React · Node.js