Michael Feil
Michael Feil is an individual machine-learning engineer and open-source creator focused on LLM inference infrastructure, CUDA/GPU performance, and making high-throughput model serving practical. His principal independent project is Infinity, an open-source serving engine for embeddings, rerankers, classifiers, and multimodal models.
At a Glance
- AI application developers
- Enterprises building private or self-hosted AI systems
- Search and retrieval teams
- RAG and vector-database builders
- +2 more
AI Tools by Michael Feil
(1)
Infinity
Embedding and Reranking Inference Server
Discussions
No discussions yet
Be the first to start a discussion about Michael Feil
Latest News
Baseten article authored by Michael Feil on 2x faster inference with NVIDIA Dynamo's KV-cache-aware routing.
Baseten announced Qwen3 Embedding support on NVIDIA Blackwell GPUs; the post lists Michael Feil as an author and reports BEI benchmarks against vLLM and TEI.
Infinity repository news listed Blackwell support.
Michael Feil authored a Baseten post introducing the open-source Rust/Python Performance Client, reporting up to 12x higher throughput for high-volume embedding workloads.
Products & Services
MIT-licensed, high-throughput, low-latency inference server and REST API for text embeddings, reranking, text classification, and multimodal models such as CLIP, CLAP, and ColPali. It supports Hugging Face/SentenceTransformers models and multiple inference backends and hardware targets.
An MIT-licensed, stable inference library for loading Hugging Face embedding, classification, and reranking models. It uses Infinity as a backend for asynchronous computation, batching, and Flash-Attention 2.
Market Position
Infinity is positioned as an open-source, self-hostable alternative for embedding and reranking inference, emphasizing low latency, dynamic batching, broad hardware/backend support, and OpenAI compatibility. Public comparisons and ecosystem references place it alongside Hugging Face Text Embeddings Inference (TEI), vLLM, Ollama, and Qdrant FastEmbed; unlike a hosted API, Infinity can be deployed directly by users on their own infrastructure.
Leadership
Founders
Michael Feil
Creator and maintainer of Infinity. He holds a bachelor's degree in mechatronics and information technology from KIT and a master's degree in Robotics, Cognition and Intelligence from Technical University of Munich. His background includes machine-learning engineering at Rohde & Schwarz, research/work with Bosch Research and the Max Planck Institute for Intelligent Systems, and LLM/inference work at Gradient and Baseten.
Founding Story
Feil started Infinity as a side project to address the difficulty of deploying SentenceTransformer and other encoder models as reliable production services. His stated goal was a fast, production-ready, open-source embedding-as-a-service framework that handled batching, serialization, parallelism, and an OpenAI-compatible API, making capable inference accessible without large infrastructure teams.
Target Markets
- AI application developers
- Enterprises building private or self-hosted AI systems
- Search and retrieval teams
- RAG and vector-database builders
- Cloud GPU and inference-platform operators
- Researchers and open-source ML engineers
- Retrieval-augmented generation (RAG)
- Semantic search and vector database ingestion
- Document clustering and classification
- Reranking search results
- Recommendation systems
- High-volume batch embedding and preprocessing
- SAP / SAP AI Core
- RunPod
- Vast.ai
- Gradient.ai