Generative AI Projects

This repository groups together my Generative AI projects in one place for cleaner portfolio management and easier exploration.

Recent additions in this repository also include content-focused, safety-focused, evaluation-focused, and agent-engineering GenAI systems:

• blog-generation-agenticai for AI-assisted long-form blog drafting and content generation workflows.
• ainews-generation-agenticai for AI-driven news generation, summarization, and topic-oriented publishing ideas.
• blood-report-parsing-iisc for AI-assisted blood report parsing, abnormality detection, and healthcare-oriented structured extraction.
• ai_guardrails for NeMo Guardrails, Colang flows, input/output rails, and safer LLM behavior design.
• agentic_memory for AI Memory Lab, a Streamlit/Groq demo suite covering short-term memory, long-term memory, vector-store memory, entity memory, episodic/semantic/procedural memory, self-reflection, routing, cost tracking, and architecture diagrams.
• advanced_prod_rag for production-ready enterprise RAG with OpenAI/Anthropic routing through Portkey, Jina embeddings/reranking, Qdrant retrieval, Prometheus metrics, auth, rate limiting, Neon/Upstash integrations, tests, and AWS deployment scripts.
• loop_engineering_demo for a TestSprite loop-engineering demo showing build -> verify -> inspect failure bundle -> fix -> deploy -> rerun around an intentional checkout regression.
• harness_engineering_demo for a LangGraph multi-agent support harness with guardrail, router, specialist agents, and reviewer agent patterns.
• google_okf for comparing Basic RAG, Google Open Knowledge Format retrieval, and Hybrid RAG with LangGraph, FAISS, Jina reranking, Groq, openevals, and LangSmith experiments.
• rag_evaluation for RAGAS-based evaluation of retrieval and generation quality in a product-catalog RAG app.
• llm_gateways for production LLM gateway concepts including Portkey routing, retries, fallbacks, observability, caching, rate limits, load balancing, and streaming.
• multimodal_rag_via_diff_ocr for comparing Docling, Unlimited-OCR/PaddleOCR, and GLM-OCR layout strategies inside a multimodal RAG pipeline with Qdrant hybrid search, RRF fusion, reranking, Groq generation, FastAPI, and Streamlit BYOK workbenches.

Included Projects

• ainews-generation-agenticai
• ai_guardrails
• agentic_memory
• advanced_rag
• advanced_prod_rag
• ats-using-gemini
• blog-generation-agenticai
• blood-report-parsing-iisc
• calorie-calc-using-gpv
• disease-diagnosis-dhanvantari
• google_okf
• harassment-bot
• harness_engineering_demo
• hybd_cmr_edtech
• loop_engineering_demo
• med-triage-agenticai
• llm_gateways
• multimodal_rag_via_diff_ocr
• pskgpt-via-transformers
• rag_evaluation
• sadhana-gen-ai-project
• stance-detection
• whatsapp-chat-analyser

Notes

• Nested Git repositories, FAISS index files, and local generated artifacts are removed so this repo can be pushed cleanly as a single grouped repository.
• Secrets and hardcoded API keys should always be replaced with environment variable usage before pushing.
• Project folders may continue to be refined over time with better documentation, screenshots, and deployment notes.