santhosh ravula

About

Software engineer who has shipped both core web platform features (secure user authentication/profile management) and production LLM systems. Built an internal documentation knowledge assistant using a full RAG pipeline (OpenAI embeddings, vector DB, semantic search, reranking) with evaluation loops and a scalable document-ingestion pipeline for PDFs/FAQs, iterating based on metrics and user feedback.

Key Strengths

• Owned end-to-end delivery of authentication and profile management (API/db/frontend/deploy)
• Balanced security vs delivery speed; shipped JWT-based auth with secure practices (hashing, token expiry, protected routes)
• Improved signup conversion by simplifying registration flow and clarifying validation/error messaging; reduced support requests
• Built production RAG knowledge assistant using embeddings + vector DB + semantic retrieval with guardrails (fallbacks, rate limiting, auth)
• Designed LLM evaluation loop (100-query test set, relevance/correctness/hallucination metrics, human review) and iterated based on results
• Improved retrieval quality by adjusting chunking strategy and adding reranking to reduce irrelevant context
• Built scalable ingestion pipeline for messy docs (PDF/FAQs/guidance) with queue-based reliability, retries, and observability
• Designed and implemented end-to-end inference pipeline (ingestion to model execution to API exposure)
• Optimized inference speed/accuracy tradeoffs with scalable architecture
• Reduced/controlled latency via timestamped logging, bottleneck identification, and replaying recorded data
• Implemented asynchronous request handling for more stable real-time response times
• Containerized services with Docker for scalable deployment on AWS
• Built ROS Python perception nodes (subscribe to image topics, run OpenCV/ML inference, publish results)

Experience

Education

Skills

Languages

Certifications