Generative AI Engineer with 4 years building production-grade AI systems across multimodal applications, Kaggle Master, author of llmevalkit & adaptive-intelligence library, passionate about advancing the frontiers of AI through innovative research that transforms data into actionable insights and intelligent systems.
I'm a Generative AI Engineer passionate about building intelligent solutions that blend research, real-world deployment, and innovation. My work spans Generative AI, Computer Vision, NLP, and end-to-end AI pipelines from designing algorithms to deploying production-ready systems.
With experience across healthcare, manufacturing, and automotive domains, I've built real-time RAG pipelines, 3D reconstruction systems using NeRF & photogrammetry, document processing platforms on Azure AI, and AI-powered automation tools. I hold an M.Tech in AI from BITS Pilani and am a Kaggle Master (Competitions Expert top ~0.4%, Notebooks Master top ~1.3%).
I enjoy bridging the gap between deep research and applied business impact with a strong focus on responsible, secure AI systems and governance.
arXiv · March 2025
IEEE · Feb 2024
IRJIET · March 2020
ResearchGate · May 2026
ResearchGate · May 2026
Universal Python library for evaluating and benchmarking LLM systems across 78 metrics in 13 modules covering Quality, Hallucination Detection, Compliance, Document AI, Governance, Security, Multimodal Evaluation, AI Content Detection, Observability, Anomaly Detection, Ground Truth Testing, Conversation Evaluation, and Red Team Testing. Works fully offline or with any LLM provider, supports RAG pipelines, AI agents, document extraction systems, table extraction workflows, and evaluation without mandatory ground-truth labels.
v6.0 Highlights: Added Ground Truth Testing, Conversation Evaluation, Red Team Testing, and Table Extraction support. Expanded the framework to 78 evaluation metrics across 13 modules with offline evaluation and LLM-as-judge support for LLMs, RAG systems, AI agents, and governance workflows.
Self-improving retrieval framework that learns, remembers, and connects tools. Uses reinforcement learning to pick the best retrieval strategy per query type. Connects external tools via MCP. Runs multi-round agentic retrieval. Optimizes the entire context window. Remembers across sessions. Works with any LLM — Ollama, NVIDIA NIM, Groq, Gemini, Claude, HuggingFace. Zero required dependencies.
v4 Highlights: Context engineering optimizes entire context window (chunks, memory, history, tools, prompt). MCP integration connects external tools and serves retrieval as MCP server. Agentic workflow runs multi-round retrieval with query refinement and tool calls. Persistent memory survives restart. Incremental learning continues when new documents added. RL policy learns retrieval routing, tool selection, and graph activation — all through one feedback loop.
Always open to new opportunities, collaborations, and meaningful conversations about AI. Reach out directly or drop a message below — I usually reply within a day.