● ACTIVEYouTube AI HelperLLM-based system that converts long-form YouTube videos and playlists into structured notes, flashcards, and a queryable interface.// Modular pipeline: transcript extraction → summarization → Q&A with multi-video context// Explored limits of full-context prompting; working toward retrieval-based approachesLLMsRAG (in progress)StreamlitPythonAgentic Workflows[GITHUB] LIVE APP
○ ARCHIVEDContent Recommendation PipelineTwo-stage recommendation system: embedding-based retrieval + local LLM reranking.// Cosine similarity retrieval paired with Llama 3B reranker on RTX 3090// Designed a category-overlap evaluation metric — then mathematically invalidated it upon finding structural embedding space boundariesRecommendation SystemsEmbeddingsLLM RerankingEvaluationPython
○ ARCHIVEDVision-to-Text SOCVision-to-text pipeline for extracting structured information from image-heavy inputs.// Combined OCR and NLP to convert visual data into usable text representations// Focused on making extracted data structured and actionable rather than raw outputOCR PipelinesDocument UnderstandingNLPPython[GITHUB]
○ ARCHIVEDWIDS 2025 Agentic AIAgentic AI system exploring multi-step reasoning and tool-based workflows.// Modular agents for task decomposition, routing, and response generation// Explored limits of agent reliability, control flow, and hallucination handlingAgent SystemsLLMsTool UsePrompt Orchestration[GITHUB]