We are a small applied AI research group that integrates with your team as a hands-on strategic partner. Our background spans Google (built Gemini) and founding teams at venture-backed startups (leading ML infrastructure, engineering, and full stack development).
Heinrich worked at Google Research for 8 years. His contributions include building the first production fact-checker within Gemini, RAG systems for Google Cloud's Vertex AI enterprise help centers, and a pre-training technique for LLMs leading to ~40% savings in compute. Heinrich has published over 30 papers at top-tier peer-reviewed conferences such as NeurIPS, ICML, ICLR, and CVPR and has over 3000 citations. Heinrich holds a bachelors degree in mathematics from Princeton University.
Henning was the first employee and Head of Engineering at Grantscout Inc., a San Francisco based VC-backed startup, where he led a team of 4 engineers doing full stack development.
We own your AI engineering efforts, including feasibility, implementation, productionization and monitoring. We integrate with your team, and we quickly turn business objectives into results without you needing to hire full time talent.
Within the first 15 business days of working with us, you will receive one of the following, depending on your needs:
If not, the first month working with us is free.
Month-to-month. Cancel anytime.
AI Research Ownership
AI Engineering Ownership
Feasibility Study
Product Implementation
Productionization
Monitoring
15 Business Day Deliverable Guarantee
Hiring Blueprint
A venture-backed startup needed a system that could answer questions about legal cases.
We built an efficient language model that identifies and extracts relevant answers from large archives. The model improved recall and significantly reduced the latency for discovery, enabling scaling to databases of thousands of documents.
A venture-backed startup helping businesses apply for government funding opportunities wanted to automate funding search.
We built an AI-powered search engine that provides a superior experience to existing government provided search engines, and the feature became an effective lead magnet for the company.
A venture-backed startup wanted an AI differentiator for designers and customers.
We built a 30B+ parameter model that turns customer ideas into precise manufacturable designs and set up a reinforcement-learning loop with domain experts to improve results over time.
Ready to talk to us? Book a with our team, email us at [email protected], or fill out the form below.