Windows AI Foundry Research
Research log on local-first AI, RAG and Copilot+ PCs. Six weeks building on Windows AI Foundry.
- Windows AI Foundry
- Foundry Local
- ONNX Runtime
- Pinecone
- RAG
- Python
- A week-by-week record of a research internship exploring Microsoft's local-first AI stack, from Foundry Local and Windows ML to Copilot+ PC hardware.
- Compared Windows AI Foundry with Azure AI Foundry and Ollama on deployment, scalability, privacy and latency.
- Built a Financial Report Generator on Foundry Local, then a full RAG application using Pinecone.
- The RAG pipeline: query rewriting, HyDE, retrieval, reranking and fusion RAG before generation.
- Benchmarked Phi-3 Mini on Foundry against Ollama and explored NPU-accelerated inference on Copilot+ PCs.