In this episode of the Sunny Ray Show, host Sunny Ray talks with Stu Frost of Gemino's about why so many enterprise AI efforts stall after an impressive demo. Frost argues that large companies are struggling to get value from copilots, vibe coding and agents because they lean too heavily on LLMs alone. Gemino's instead combines causal AI, knowledge graphs, business rules, conventional code, LLMs and agents to build decision making applications quickly and reliably. Frost explains causal AI in CFO terms, describes how LLMs read text and feed a causal model that makes the final call, and positions his curated knowledge graph approach between model company automation and Palantir style ontologies. He walks through a Great Western Railways delay attribution project that went from a few days of demo work to a full system in about eight weeks, with humans staying in the loop. He also covers land and expand sales, shorter sales cycles after shifting to selling solutions, angel funding, a planned Series A, and what convinces skeptical boards.
Why enterprise AI demos stall, and how pairing causal models with knowledge graphs gets real decisions into production in weeks.
What problem is Gemino's solving?
Big companies struggle to get value from AI. Copilots and vibe coding produce glossy demos that are hard to repeat or trust. We combine causal AI, knowledge graphs, business rules, regular code, LLMs and agents to build specific decision making applications fast, reliably and at scale, rather than depending on LLMs alone.
Can you explain causal AI through a business decision a CFO would recognize?
We pick a key outcome like profitability, then map its drivers, such as sales and costs, in terms a CFO understands. A subject matter expert checks the model, then we add data and build the math. That gives a repeatable engine to test interventions and fix outcomes that drift.
How do LLMs, the knowledge graph and the causal model work together?
LLMs understand text, voice and messages. The knowledge graph, built from the customer's documents, tells them what those words mean for that specific business. We extract the relevant information and feed it into the causal model, which makes the data driven decision. LLMs help understand, but they don't decide.
Aren't the big model companies already doing knowledge graphs for enterprise?
Their approaches are failing often, with something like an 87% failure rate going from demo to production. Graph RAG is mostly machine readable, but enterprises need curation. Model companies automate everything, while Palantir spends millions on ontologies. We sit in the middle, with real depth but built in days.
Can you walk us through a customer problem and how it moved from demo to deployment?
Great Western Railways needed to attribute train delays, which took eight to nine days, so they missed recovering money. In a few days we showed a solution using synthetic data and agents. They signed, and about eight weeks later we delivered. Humans make the final decision, now in minutes.
What changes for the people doing the work?
The tedium goes away. We capture best practices built up over decades and share them across the team. Instead of pulling data from 15 sources, people see probabilities, like 70% one cause and 20% another, ask questions in natural language, and focus on analysis and decisions.
What did switching from selling tools to selling solutions change?
We were building solutions to prove our tools worked, and realized customers wanted help making better decisions. About seven or eight months ago we switched to selling solutions to the business side. Sales cycles dropped from nine to twelve months down to one or two, before procurement.
What convinces a skeptical board that AI is infrastructure and not a science project?
Delivering value quickly. If the first use case or two make a decision more consistent and much faster, that is very convincing. We build these for about $100,000, sometimes a bit more, with a reasonable monthly license fee, so the barrier to entry is low.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interviewSend a founder here and we will know it was you.