CASSANDRA: The Failure Prophet
Every year, the tech industry builds a monument to innovation. But at its feet lies a graveyard of forgotten startups, filled with brilliant ideas and tireless teams that simply made the wrong turn at the wrong time. The lessons from this graveyard are the most valuable, most expensive dataset in the world. But it's locked away, unstructured and unheard.
Our inspiration was to build a gatekeeper for that graveyard. An oracle.
We built Cassandra, a "pre-mortem co-pilot" whose only job is to learn from the dead to guide the living. We didn't want to build another RAG demo that fetches documents. We wanted to build an agent that could provide genuine foresight, saving founders their time, their capital, and their sanity by turning that graveyard of failures into a strategic map.
The Journey: From an Idea to an Insight Engine
A founder gives Cassandra a pitch—an idea, a business plan, a new feature. Instantly, our agent awakens and begins a multi-step journey.
It doesn't just search; it investigates. It analyzes the idea, then dives into its knowledge base of over 1,000 startup post-mortems. But here, we hit a wall. We realized that failure is nuanced. A "market fit" problem can be semantic—a fundamentally flawed idea—or factual—a dozen competitors already exist. A simple database, whether keyword or vector, is half-blind. It can't see both.
So, we gave our agent a unique kind of mind, powered by TiDB Serverless. It has two halves of a brain, working in perfect concert:
- The Conceptual Brain (Vector Search): One half understands analogy and abstraction. It uses
VEC_COSINE_DISTANCEto find startups that failed for the same reasons, even if they were in totally different markets. - The Factual Brain (Full-Text Search): The other half is a master of detail. It uses full-text search to find startups that failed in the same market, facing the same direct competitors and conditions.
This hybrid search is our unfair advantage. It allows Cassandra to deliver an analysis with a level of nuance and accuracy that a single-mode database simply cannot achieve.
But an insight is useless if it's not understood. We rejected the soulless, linear chat window. Instead, we give founders what they actually need: a map.
Our UI, built on React Flow, is a strategic canvas. The AI's analysis becomes the starting point of an explorable "Failure Map." A founder can visually trace lines of risk, drill down into a specific threat to spawn a new line of inquiry, or merge disparate insights to synthesize a new, higher-level strategy. It transforms a passive conversation into an active, strategic workspace.
Our Vision
This isn't just a hackathon project. It's the foundation for a new class of strategic tool. We envision a future where Cassandra integrates with project management tools to flag at-risk features, where VCs use it to spot systemic risks across their portfolios, and where every founder has a co-pilot armed with the hard-won wisdom of a thousand failures.
We built Cassandra to prevent preventable failures. By giving founders the gift of foresight, we believe we can change how the world builds.


Log in or sign up for Devpost to join the conversation.