News Reporter: A Multi-lingual LLM Framework for Broadcast T.V News
A specialized multi-lingual LLM framework for broadcast television news with transcript QA dataset curation, model fine-tuning, and RAG integration for verifiable news grounding.
Research & Talks
Research, publications, conference talks, and workshops focused on retrieval, evaluation, and applied AI systems.
Publications & Research
Talks & Workshops
Talk and mini workshop on debugging LLM applications with Opik traces, spans, evaluations, datasets, and feedback loops.
Talk on relationship-aware long-term memory for AI Agents, combining semantic search and structured graph traversal for multi-hop reasoning.
Conference session on long-term memory patterns for Agentic workflows at AgentCon Nairobi.
Workshop on building a completely local Agentic RAG system using Gemma/Mistral, Qdrant vector search, and Comet OPIK - no external APIs.
Keynote session in the Main Keynote Hall on persistent memory architectures for AI Agent workflows at Google's Builder Day.
Three-hour workshop by Consult Club and Analytics Club at IIT Bombay.
Deep-dive into Qdrant Hybrid Search, Maximum Marginal Relevance (MMR), and vector scoring.
Architecting intelligent retrieval routing mechanisms with Agentic workflows.
Hands-on workshop with Model Context Protocol to feed rich context into AI Agents.
Conducted an end-to-end workshop building RAG pipelines using open-source tools.
Delivered talk at Google RMZ Office on building production Agentic retrieval pipelines.
Keynote on parameter-efficient fine-tuning and retrieval augmented generation.