Compression at the Edge
Chris Alexiuk · Daniel Han · Asma Beevi · Merve Noyan · Parth Sareen
AI Engineer World's Fair 2026 · 46:01
AI development platform and open-source tooling
Hugging Face builds tools and infrastructure for researchers, developers and enterprises to share and use machine learning. The Hugging Face Hub hosts models and datasets, while Spaces lets developers publish interactive AI applications. Inference Providers offers model access through a unified API; Inference Endpoints provides managed deployment. Paid team and enterprise services add access controls, security and support. Its 2025 acquisition of Pollen Robotics extended its LeRobot software work into hardware, including Reachy 2 robots for research, education and embodied AI experiments.
Founded in 2016 by CEO Clément Delangue, Julien Chaumond and Thomas Wolf, the company began with a consumer chatbot before pivoting to open-source machine learning infrastructure in 2019. Its Transformers library brings pretrained models and model architectures together under a unified API, helping developers reuse implementations across tasks. Another engineering contribution, Safetensors, stores model weights as numerical data rather than executable Python pickle content and supports memory-mapped loading.
In its summer 2026 analysis, the company reported 2.96 million public model repositories, one million datasets and 1.44 million Spaces. Its reported annualized revenue run rate reached $150 million in August 2026. A 2023 financing raised $235 million at a $4.5 billion post-money valuation. On August 26, 2026, Nvidia reportedly agreed to acquire Hugging Face for $12.9 billion; neither company had confirmed the agreement in that report.
Chris Alexiuk · Daniel Han · Asma Beevi · Merve Noyan · Parth Sareen
AI Engineer World's Fair 2026 · 46:01
AI Engineer World's Fair 2026 · 20:37
AI Engineer World's Fair 2026 · 21:39
AI Engineer World's Fair 2026 · 17:28
AI Engineer Europe 2026 · 21:16
AI Engineer Europe 2026 · 19:11
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here to learn how CUDA kernel work, prompt-driven fine-tuning, and coordinated research experiments place different demands on coding agents.
Ben BurtenshawAI Engineer Europe 2026
Read for a concrete infrastructure example connecting MongoDB Atlas, AWS S3, Atlas Search, and Kubernetes autoscaling.
Arek BoruckiAI Engineer World's Fair 2026
Start here for Reachy Mini’s hardware configurations, multimodal demonstrations, and the reported latency and synthesis speed of FasterQwenTTS.
Andres MarafiotiAI Engineer Europe 2026
Hugging Face engineer Niels Rogge explains how he automates research-community outreach that encourages authors to migrate models and datasets from scattered hosting services to the Hugging Face Hub.
Niels RoggeAI Engineer World's Fair 2026
Niels Rogge contrasts a predefined outreach workflow with an autonomous follow-up agent. Ben Burtenshaw and Merve Noyan describe skills, training tools, and experiment workflows that expand what agents can execute.
Arek Borucki describes prioritizing P99 search latency through separate metadata and artifact storage, denormalized read collections, and tokenized autocomplete with relevance ranking.
Andres Marafioti connects conversational robot interactions with speech synthesis optimization. The compression panel featuring Merve Noyan discusses model-size versus quality tradeoffs and evaluation for local and edge deployment.
Affiliations reflect each recorded session, not necessarily current employment.