AXIOM AI-native Autonomous Growth Operating System

Inspiration We were inspired by a critical flaw in the modern B2B SaaS landscape: teams are drowning in data but starving for action. Standard analytics tools are excellent at reporting what happened, but they fail to explain why it happened or what to do next. When Google DeepMind announced the Gemini 3 Hackathon with a focus on "enhanced reasoning capabilities" and "more than just another chat interface", we saw the perfect opportunity. We wanted to move growth from a manual, reactive task to an automated, predictive utility—building an operating system that doesn't just read charts, but actively intervenes to fix them. What it does AXIOM is an AI-native Autonomous Growth Operating System. It centralizes fragmented growth processes—analytics, experimentation, and execution—into a single intelligent fabric. Unlike a standard dashboard, Axiom operates as an agentic loop:

1. Ingestion (The Unified Growth Fabric): Unifies siloed data from CRMs and ad platforms into a Predictive Growth Graph.
2. Analysis (Causal Inference): Uses a Causal Inference Engine (Double Machine Learning) to distinguish between correlation and causation.
3. Optimization (Autonomous Execution): Uses an Autonomous Control Room (Multi-Armed Bandits) to dynamically route traffic to winning experiment variants.
4. Simulation (Ghost Users): Deploys "Synthetic Agents" to navigate the website and identify friction points before real humans encounter them. How we built it We built AXIOM as a new application specifically for this hackathon, utilizing the Gemini 3 API as the core intelligence layer. • Reasoning Core: We utilized Gemini 3's reasoning to implement Double Machine Learning (DoubleML), allowing us to estimate causal parameters ($ \theta_0 $) and identify confounders in unstructured data. • Autonomous Execution: We implemented Thompson Sampling algorithms where Gemini 3 analyzes performance and adjusts the probability of serving specific variants. • Multimodal & Latency: We leveraged Gemini 3's multimodal capabilities for the "Generative Copy Studio" (creating images/emails) and its reduced latency for "Axiom Live," a real-time voice interface. Challenges we ran into • The "Public Access" Constraint: Since AXIOM simulates an enterprise tool requiring sensitive API keys, we couldn't use real user data. We had to build a robust "Mock Data Layer" that simulates a B2B SaaS environment, allowing judges to experience the platform without credentials. • Hallucination in Logic: Early versions of the Journey Builder suggested mathematically impossible strategies. We solved this by using Gemini 3’s reasoning to cross-reference generated strategies against our Causal Inference Engine. Accomplishments that we're proud of • Agentic Workforce: Watching a Gemini-powered "Ghost User" autonomously navigate a mock website and report friction points is a huge technical milestone. • Visualizing Causality: Successfully rendering the force-directed Predictive Growth Graph makes complex causal math intuitive. What's next for AXIOM We plan to expand the "Strategic War Games" module, allowing AI agents to play the role of competitors to stress-test business strategies.## Inspiration

Built With

• causal-inference
• double-ml
• framework)
• frontend
• google-gemini-3
• python
• react
• thompson-sampling
• your