• GenAI Engineer at Tredence Inc. (Sept 2024 – Present), building RAG and agentic AI systems for enterprise clients.
• B.Tech, Metallurgical & Materials Engineering, NIT Karnataka (Surathkal), 2020–2024.
• Core focus: hybrid retrieval fusion (BM25 + dense + RRF), agentic orchestration with LangGraph, and evaluation/trust — confidence scoring, hallucination detection, ablation studies.
• Shipped a GenAI solution to production on the Snowflake Apps platform, beyond its original proof-of-concept scope — a deployment signal, not just a demo.
• Industry exposure: medical devices (regulated document Q&A) and medtech supply chain (commercial root-cause analysis for inventory obsolescence).
• Certified: Databricks Generative AI Engineer Associate, Microsoft Azure AI Engineer Associate (AI-102), Databricks Machine Learning Associate.
• Interested in evaluation-driven AI engineering — systems that are only trusted once they're measured, not just built.

github.com/charan-ravikumar/Advanced-Rag

A hybrid RAG pipeline (spaCy semantic chunking, BM25 + dense retrieval, RRF fusion, optional CrossEncoder reranker) evaluated through a rigorous 4-config ablation study, shipped as a real service — not just a notebook.

• Stack: Python, FastAPI, spaCy, ChromaDB Cloud, rank_bm25, sentence-transformers, RAGAS, Redis, Prometheus
• Impact: Improved Recall@5 by 43.6% (0.188 → 0.270) while cutting P50 latency to 1.7s. Isolated the actual source of the gain: RRF fusion alone, not BM25 in isolation.
• Engineering challenge: Designing four ablation configs as pure parameter variants (not code variants) to rule out implementation-difference confounds, then backing every claim with a custom evaluation harness (Recall@K, MRR, P50/P95/P99 latency) plus an optional RAGAS layer.

github.com/charan-ravikumar/Multi-Agent-AI-Research-Analyst

An autonomous LangGraph research system — planner, parallel research agents, a bounded self-critique loop, and a human-in-the-loop approval checkpoint — that discloses contradictions and gaps instead of hiding them.

• Stack: Python, LangGraph, Groq, Gemini, Redis, Streamlit
• Impact: Built an LLM-judged evaluation harness that caught the self-correction loop actively degrading faithfulness (93.3% → 39.1%); root-caused it to an ambiguous prompt, fixed it, and verified recovery to zero genuine hallucination.
• Engineering challenge: Proving the self-correction loop was structurally correct (bounded, terminating) was not enough — the eval harness was what proved it was actually harmful, then confirmed the fix worked. Verification over assumption, end to end.

• Production deployment beyond PoC scope — delivered a Snowflake Cortex GenAI solution on the Snowflake Apps platform, where the original engagement called for a non-executing proof-of-concept.
• Regulated-domain RAG — designed retrieval, confidence scoring, and query routing for a medical-device document-intelligence platform (FDA / ISO 13485 / ISO 11608 context), where a confidently wrong answer is worse than no answer.
• Enterprise-scale retrieval — RAG corpus of 250+ enterprise documents with a three-tier hybrid fusion architecture and a two-signal confidence-scoring engine (retrieval similarity + LLM-based claim grounding).
• Agentic systems with real evaluation — LangGraph multi-agent orchestration with bounded self-correction loops and human-in-the-loop checkpoints, validated with an LLM-judged evaluation harness rather than assumed to work.
• Real-time computer vision at scale — parallel multi-model inference pipeline cutting per-person latency ~8x (~400ms → ~50ms), running at ~25 FPS on CPU-only hardware.

• Deepening evaluation methodology for agentic systems — moving beyond "does it run" to "does it actually help," the way the faithfulness regression in the Multi-Agent Research Analyst project was caught and fixed.
• Exploring GPU-backed reranking and lower-latency hybrid retrieval configurations.
• Strengthening statistical/EDA fundamentals to complement evaluation-methodology work.

• Hybrid retrieval and RAG evaluation tooling
• Agentic workflows built on LangGraph with human-in-the-loop design
• Production-oriented GenAI systems (observability, caching, deployment — not just prototypes)

Feel free to open an issue or reach out directly.

• Databricks Certified Generative AI Engineer Associate
• Microsoft Certified: Azure AI Engineer Associate (AI-102)
• Databricks Certified Machine Learning Associate

Bengaluru, India