Large-scale pretrained models for goal-directed dialog

Open-vocabulary Object Segmentation with Diffusion Models

This is a list of open-source projects at Microsoft Research NLP Group

Financial report grounded QA post-training with QLoRA SFT, DPO preference learning, citation/numeric evaluation, and badcase analysis.

Beginner-friendly Python RAG project that turns StackOverflow Q&A data into a grounded knowledge assistant using sentence-transformer embeddings, FAISS retrieval, and strict citation-based LLM answering.

Developer-first prompt engineering patterns for grounded, testable, and reliable AI outputs.

Secure Enterprise RAG platform with RBAC enforcement, OWASP LLM defenses, hybrid retrieval, grounded generation, explainability, and zero-trust context pipelines.

Turn a product idea into a grounded, buildable SRD suite — a skills.sh agent skill

Grounded RAG v1 specification with abstain/partial semantics

Adaptive quiz generation from PDFs using interpretable NLP, grounded LLMs, and deep knowledge tracing.

URIM is a local grounded chat runtime for querying boxed authority content with a fixed, deterministic retrieval pipeline.

KG-based Text World Models & Semantic Dynamics for Conversational AI — build, simulate, and reason over a user's semantic world with text-native graphs and time-aware rules.

ComfyUI Custom Nodes for Gen Searcher and other tools

Production-oriented local RAG system with FAISS, FastAPI, and Docker, featuring confidence-gated generation, safe refusals, and grounded citations. Built with real-world reliability and deployment in mind.

Private document intelligence platform with grounded RAG, hybrid retrieval, reranking, structured chunking, local LLMs, answer verification, RBAC, and observability.

RAG system for Trine University international students. Provides grounded answers on admissions, CPT/OPT, scholarships, insurance using 10 FAQ documents. Built with Python, ChromaDB, Groq LLM, and Gradio UI

Personal RAG avatar agent that answers in my voice using a private, continuously evolving knowledge base with grounded retrieval.

Strictly-grounded RAG chat on Adobe Spectrum docs — Algolia Agent Studio + Algolia index. Per-answer grounding judge (3-judge panel: Skeptic/Referee/Advocate) surfaced as a Confidence chip → breakdown drawer. Reusable Algolia-Central-[company] template.

Anchored {{key | value}} placeholders for hallucination-safe LLM generation in finance.