Raahul Krishna Durairaju

About

Built and deployed real-world systems end-to-end across security and healthcare contexts: led a 3-person team delivering a university vehicle tracking system with 30% cost savings and 1-year post-launch monitoring. Also implemented a healthcare RAG chatbot with adaptive query routing that cut LLM costs by 40% while maintaining answer accuracy, and has experience debugging non-deterministic LLM behavior in DevOps pipeline automation.

Key Strengths

• Owned end-to-end deployment of a university vehicle tracking/security system (discovery through 1-year post-launch monitoring)
• Led a 3-person team and managed stakeholders across university administration and campus police
• Delivered measurable impact: 30% cost savings vs manual vehicle entry and improved campus security
• Resolved complex integration issues (hardware/driver mismatches) by modularizing deployment and prioritizing critical components
• Built and deployed a healthcare RAG chatbot with strong information retrieval design (chunking + vector store)
• Reduced LLM operating costs by 40% via adaptive intent-based query routing to different model tiers
• Systematic production debugging by isolating pipeline steps to identify bottlenecks in non-deterministic LLM-driven DevOps workflows
• Shipped production LLM agent automating CI/CD via Slack with retries, recovery, approvals, and observability
• Reduced manual DevOps intervention ~60% and improved pipeline turnaround time
• Architected multi-agent fact verification pipeline processing ~50,000 news articles/day
• Implemented schema-validated structured outputs, confidence thresholds, and anomaly monitoring for predictable agent behavior
• Reduced manual fact checking by over 90% via automated claim verification and publishing
• Designed offline/online evaluation loops (precision/recall/NDCG + Grafana engagement/CTR) and iterated based on error analysis
• Improved recommendations for niche genres via training data oversampling, metadata enrichment, and embedding/regularization tuning
• Handled messy multi-source data with pandas normalization, automated schema validation, idempotent updates, and RBAC/permission checks

Experience

Education

Skills

Languages

Certifications

Publications

1 publication