Want to know if this job is worth applying to?
San Francisco, California, United States
Hybrid
From humanoid robots to autonomous vehicles, every Physical AI model is trained on petabytes of video, lidar, radar, and sensor data. Today’s data platforms (Databricks, Snowflake) were built for spreadsheet-like analytics, not video corpora. And understanding that video still means paying a person to watch it, ten dollars an hour of footage at the low end. So teams check a sample and hope it represents the rest. The footage grows every year; the budget to look at it doesn’t.
Eventual was founded in 2022 to close that gap. Our open-source engine, Daft, is purpose-built for multimodal AI: 2 PB/day at Amazon, 60-100 PB at another FAANG company, and in production at companies like Mobileye, TogetherAI. On top of it we’re building the infrastructure that finds any situation you can describe across a fleet’s entire video history, and turns it into a training set or an alert someone can still act on. We fine-tune and run the vision models ourselves, which makes indexing every hour cheaper than annotating a sample.
We’re building this with the top Physical AI labs and GPU cloud providers. We’ve raised $30M from investors like Felicis, CRV, Y Combinator, and angels from the co-founders of Databricks and Perplexity. Our team comes from AWS, Lyft, and Tesla. We powered the last generation of Physical AI in self-driving; now we’re doing it for the next.
Join our small (but powerful!) team, 4 days/week in our SF Mission District office.
As Technical Lead, Multimodal Research, you’ll own the execution of our technical vision behind everything Eventual can understand about a video. Physical AI teams have video, lidar, radar, and sim outputs scattered across object stores with no way to find what they need without weeks of human annotation. Eventual runs vision/language models and pipelines over every clip in a corpus along axes the customer cares about (gripper type, failure mode, object class, scene, motion density), so a researcher can ask “left-arm grasp failures on deformable objects” and get a curated dataset in minutes.
You’ll decide which models, representations, and evaluation methods get us there, and prove them in production at petabyte scale, over hundreds of thousands of hours of video. This is a senior individual contributor role rather than a management one: you set research direction and make the architectural decisions, while staying hands-on with papers, models, and experiments.
Own modeling strategy across the platform rather than one customer’s taxonomy: which model families, representations, and training approaches we invest in, which get prototyped, and when to move off one.
Take approaches from prototype into production inference at corpus scale, working with the data systems and storage teams on what the index and the loader require.
Define the evaluation standard — the benchmarks and acceptance criteria any model meets before it reaches a customer.
Own the cost curve for understanding: architecture-level decisions on distillation, cascades, routing, and quantization that keep a 10K-hour corpus at single-digit cents per hour of video.
Translate customer research needs into scoped technical programs — taxonomy, model plan, datasets, quality instrumentation — and set the technical direction for multimodal work across the company. No direct reports.
5+ years in applied computer vision or multimodal ML.
PhD or MS in computer science, electrical engineering, robotics, or applied mathematics with a computer vision or machine learning focus. A comparable publication or production record is acceptable in place of the degree.
Depth in modern vision and multimodal modeling — VLMs, VQA, embeddings, representation learning, detection, tracking, segmentation, retrieval — with judgment about what is deployable today rather than competitive on a leaderboard.
Hands-on training and evaluation of these models at scale on real video and sensor data, and comfort across the research and engineering boundary: PyTorch prototyping alongside inference performance, GPU utilization, throughput, and cost.
Background from a perception or multimodal team at a self-driving, robotics, or Physical AI company, a frontier research lab, or a visual-data company, ideally as the senior-most person on that problem.
Publications at CVPR, ICCV, ECCV, NeurIPS, ICML, or ICLR.
Built or fine-tuned VLMs or other multimodal foundation models.
Long-form video, temporal reasoning, embeddings, retrieval, or content-aware indexing at scale.
Multimodal sensor data beyond RGB — lidar, radar, depth, or simulation output.
Evaluation frameworks, labeling taxonomies, or large-scale annotation programs, or inference and training optimization across large GPU clusters.
In-person, tight-knit team — 4 days/week in our SF Mission office.
Competitive comp and meaningful startup equity.
Catered lunches and dinners for SF employees.
Commuter benefit.
Team-building events and poker nights.
Health, vision, and dental coverage.
Flexible PTO.
Latest Apple equipment.
401(k) plan with match.
