Eventual
Eventual writes about data curation for Physical AI, and about Daft, our open-source data engine.
- Posts
- 40 from 6 Jun 2023 to 22 Sep 2026
- Posting
- monthly (23 in the last year)
- Kind
- company blog
2026 17
- open sourceSeptember 22, 2026What I learned re-evaluating VLADBench on the latest VLMsGPT-6 Astra and Claude Opus 5.5 top the leaderboard, but the cost-performance frontier is a more interesting story.
- companyMay 5, 2026What is physical AI - is it more than just hype?Physical AI has become a real trend, but is there something real here or is it just hype?
- companyMay 8, 2026VLAs are dead, long live World Action Models - a summary of Jim Fan's Robotics End Game talkJim Fan argues robotics will follow the exact LLM playbook - and VLAs are already being replaced by World Action Models.
- open sourceMay 13, 2026Scaling As-of JoinsHow we built, broke, and re-built our ASOF joins — 5.5x faster, half the memory of pandas, and scaled to a distributed cluster.
- companyJune 2, 2026The limiting factor in physical AI isn't compute or architecture - it's dataRobotics is hitting a data wall. The architecture debate gets much of the attention, but the data constraint is more fundamental.
- open sourceJune 5, 2026Disk is the data plane: Flight Shuffle in DaftHow Daft rebuilt distributed shuffle around Arrow Flight, local disk, and streaming reads to handle multi-terabyte workloads.
- open sourceJune 26, 2026Finding a Needle in the Haystack: Querying Physical AI Data with DaftPose + semantic search over Apple's EgoDex hand-manipulation dataset with Daft: SigLIP embeddings meet hand-pose geometry. Ctrl+F for physical AI data.
- open sourceJuly 16, 2026How we made our LeRobot video reader up to 15× fasterLeRobot has emerged as the dominant open format for robot learning data, but decoding frames is expensive. Here's how we made Daft's native LeRobot reader up to 15× faster.
- companyJuly 29, 2026What if more data alone won't solve robotics? Ken Goldberg on why good old-fashioned engineering still mattersKen Goldberg argues that robotics needs more than data - modularity, algorithms, and metrics still matter. A breakdown of his MIT talk on the 100,000 year data gap.
- open sourceAugust 3, 2026Cutting dead frames from DROID: 500 hours of robot data in 32 secondsThe robot state recorded next to the video tells you which frames matter before you decode any pixels. Trimming all of DROID takes 32 seconds on a laptop.
- Image Embeddings: Tutorial & Examples
- companyOctober 14, 2026 · Everett KlevenWhat is hindsight labeling?Hindsight labeling turns recorded robot episodes into labeled data for training downstream. Its history from HER to Gemini 4 Argon and data curation.
- Multimodal Embeddings: Tutorial & Examples
- Open-sourcing 43 Billion Tokens of SEC EDGAR
- Tuning Daft's Distributed UDFs: Lessons from ByteDance
- Why I joined Eventual
- Introducing Dynamic Batching: Auto-Tuning for Daft Pipelines
2025 20
- How We Use AI Coding Agents
- Why I Joined Eventual
- Multimodal Structured Outputs: Evaluating VLM Image Understanding at Scale
- Processing 99% of U.S. Caselaw for Under $1 in the Common Pile
- Prompting with DataFrames: Massively Parallel LLM Generation is Here
- Cutting LLM Batch Inference Time in Half: Dynamic Prefix Bucketing at Scale
- Simplifying Voice AI Analytics with Daft: Transcription, Summaries, and Embeddings at Scale
- Using PyTorch DataLoaders to Streamline Multimodal Data
- Benchmarks for Multimodal AI: Spark, Ray Data, and Daft
- Introducing Flotilla: Simplifying Multimodal Data Processing at Scale
- Exploring Daft's Local Execution: The Swordfish Engine
- After the First Run
- Making GPUs Zoom (Part 1)
- End-to-End Distributed PDF Processing Pipeline
- 24 Trillion Tokens, 0 Crashes: How Essential AI Built Essential-Web v1.0 with Daft
- Embedding Millions of Text Documents With Qwen3
- Processing 300K Images Without OOM
- Eventual Raises $30M to Build the Future of Data Processing
- We cloned over 15,000 repos to find the best developers
- High-Performance File System Support With DeepSeek 3FS