The Venture Codex

Participated · Series A · Jul 2025

Vellum provides an end-to-end platform that lets cross-functional teams define, test, deploy and monitor AI workflows through a visual UI and SDK. The platform couples workflow orchestration, evaluations, testing, deployment and live monitoring to help teams move models from prototype to production. Vellum is GA and emphasizes reliability, compliance, and observability, making it suited for highly regulated industries. Customers such as Drata, Swisscom, Redfin and Headspace use the platform for use cases from automated customer support to secure, compliance-ready workflows. The company reports Vellum customers achieve 10x faster time to market and cites broad enterprise demand where many organizations lack comprehensive AI strategies. Vellum is venture-backed and has received $24.5 million of funding to date. Vellum.ai builds tooling to improve generative AI prompting, born from founders' experience building production LLM applications. The product includes side-by-side model output comparison, company-specific data search to add context to prompts, testing, version control and fine-tuning/semantic search capabilities. The team—led by CEO and co-founder Akash Sharma with co-founders Noa Flaherty and Sidd Seethepalli—spun out of work at Dover and surfaced at Y Combinator’s Winter 2023 demo day. Vellum positions itself to help companies take LLM prototypes into production by preventing hallucinations and handling edge cases through refined prompts and prompt-chaining logic. The startup reports 40 paying customers and revenue growing roughly 25%–30% month-over-month, with pricing in the three- to four-figure-per-month range and a healthy seed-stage run rate. The founders argue the expanding enterprise demand for LLMs will drive continued need for prompt-engineering tooling.

Participated · Series A · Oct 2024

Genmo’s flagship product is Mochi 1, an open-source text-to-video model released as a research preview under the Apache 2.0 license. Mochi 1 is a 10-billion-parameter model built on Genmo’s Asymmetric Diffusion Transformer (AsymmDiT) architecture and uses a video VAE to compress data; the company publishes full weights and code on Hugging Face and offers a hosted playground. The current preview supports 480p output and claims strong prompt adherence and high-fidelity motion, with a Mochi 1 HD (720p) version expected later this year. Running the full model locally requires at least four Nvidia H100 GPUs. Genmo emphasizes motion quality and plans to add image-to-video synthesis and improved controllability, with a long-term vision of powering simulators for robotics and autonomous systems. The company is actively hiring researchers and engineers as it advances the model and ecosystem.

Participated · Series A · Sep 2022

InsightFinder AI builds an AI observability platform that detects, diagnoses, remediates, and helps prevent failures across data, models, and infrastructure. Its newest product, Autonomous Reliability Insights, combines unsupervised machine learning, proprietary large and small language models, predictive AI, and causal inference in a data-agnostic base layer that ingests and correlates full data streams to arrive at root causes. The company traces its origins to 15 years of academic research and has been applying ML to infrastructure monitoring since 2016 under founder and CEO Helen Gu. InsightFinder serves enterprise customers including UBS, NBCUniversal, Lenovo, Dell, Google Cloud, and Comcast and reports revenue that has grown over threefold in the past year. The team is currently fewer than 30 people and plans to hire its first dedicated sales and marketing staff as part of its go-to-market expansion. It competes with incumbents and peers such as Grafana Labs, Fiddler, Datadog, Dynatrace, New Relic, and BigPanda.