←
Fatos20,861 posts
Fatos avatar

Fatos

@bytebiscuit

Founder of #DataScienceInitiative. #ML / #MLOps / #AI / #Data @Microsoft. Prev. Cloud Architect @Oracle. Opinions are my own.

◷ Joined February 2011▧ 8,931 Photos and videos
Microsoft Research avatar
⟳ Fatos retweeted
Microsoft Research@MSFTResearch·Oct 21

AI agents are writing more code. How can developers ensure that code is correct? Microsoft Research’s Proof-Driven Development (PDD) Kit explores a more rigorous approach: it translates a developer’s intent into a formal specification, then uses formal verification to prove that generated code meets it. PDD Kit works with Verus for Rust and is integrated with @GitHubNext’s Chopin, bringing formal methods into an AI-assisted development workflow for teams. See the work in action at @github Universe, Oct. 28–29, 2026.

0:34
Fatos avatar
Fatos@bytebiscuit·Aug 28

Please let's dispel the myth that taking AI applications to production is just an LLM call away. As this picture shows it is much more complicated than that and often this is overlooked.

Post attachment
Fatos avatar
Fatos@bytebiscuit·Aug 26

Part 5: How RAPTOR RAG works: RAPTOR RAG is built for the bigger picture. Instead of retrieving isolated chunks, RAPTOR organises knowledge into a hierarchy: Chunks → Clusters → Summaries → Tree This lets RAG retrieve at different levels of detail, from a specific fact to a broader theme spanning multiple documents. 🌳 Search across levels 🔗 Connect information across chunks 📚 Better for long, complex documents 🎯 Retrieve at the right level of detail Think of it as RAG that can see both the leaves and the tree.

0:05
Fatos avatar
Fatos@bytebiscuit·Aug 26

Part 3: How Rerank RAG works Rerank RAG is about quality over quantity. Retrieval finds the most similar documents. But the most similar documents aren’t always the most relevant. That’s where reranking comes in. Query → Retrieve Top-K → Rerank → Best Context → LLM The idea is simple: 🔎 Retrieve broadly for recall 🎯 Rerank for precision 🧠 Give the LLM only the best context Instead of stuffing more documents into the context window, reranking helps decide which documents actually deserve to be there. Better ranking → cleaner context → better answers.

0:04