We are a small team of operators, researchers, and engineers building the data infrastructure layer for enterprise AI.
The founders grew up building technical projects together. Ten years of shared reps, known failure modes, and communication speed have built the history that lets a small team ship complex systems under startup pressure.
Dilan owns DELTA agent orchestration. David owns the ML pipelines that turn raw multimodal data into training-ready datasets. Sahil owns GAMMA, the dataset specification engine. Brennan owns strategy, capital, and market. The entire stack is built internally.
Brennan analyzed hundreds of AI infrastructure companies at DTCP, NFX, and Notable Capital, mapping competitive landscapes and benchmarking architectures. The bottleneck he kept seeing is exactly what Katalyxt solves: efficiently generating structured, domain-specific training data from raw unstructured input.
Dilan has architected Level 4 agentic systems, leads LLM post-training research at UCLA, delivered 75% throughput on DoD C2 software, and developed counterfeit-detection vision models deployed on 1B+ products. David has built cortical organoid pipelines at UCSD, shipped embedded firmware at General Atomics, and developed ROS2 autonomous vehicle stacks. Sahil has built async options pipelines at ATOMA and deployed NLP models ranking over 500 startups at Karmel Capital. Brennan has authored investment theses on AI-SOC and agent infrastructure.
Freshman year of high school, the four of us founded EcoZ, a climate-tech startup making biodegradable plastic by replacing polyester with cornstarch. We went through the GATSVI incubator, took the Gold Prize, closed pre-seed funding, and sold biodegradable straws door-to-door to restaurants in our neighborhood. Covid shut operations down a month into revenue.
The company didn't compound. The reps did. Six years later, we know exactly how each of us performs under pressure and where each person's judgment is worth deferring to. That kind of trust only shows up after you've already built something together.
The AI industry is approaching a data bottleneck. As models improve, the limiting factor becomes access to high-quality training data rather than compute or architecture. Solving it requires a team that builds both the agents that generate datasets and the infrastructure that processes raw data at scale. That is the team.
