yosemite

Compliance intelligence for small businesses. Sanctions screening, anomaly detection, and geopolitical risk monitoring — powered by AI, built for the businesses that can't afford it.

Comment

Project Submission: yosemite 🏔️

Project Inspiration ⚖️

2026 has been a turbulent year for the economy worldwide, especially for the everyman impacted by global powers and politics.

Small businesses without dedicated legal teams or financial analysts are disproportionately vulnerable to fraud and trade sanctions. Even minor incidents such as a single fraudulent chargeback or an unintentional transaction with a sanctioned entity, which can be devastating to their bottom line and reputation.

With fraud on the rise and recent geopolitical conflicts causing volatile prices and frequent supply chain disruptions, we built yosemite 🏔️ to empower small businesses. Our mission is to provide an accessible, state of the art tool that allows businesses to analyze and navigate these complex risks on their own, levelling the playing field against larger enterprises with massive compliance budgets.

Technology Stack 🛠️

The project is built with a focus on high performance, safety, and modern AI capabilities.

Languages: : 🦀 Rust (Backend), 🐍 Python (ML), ⌨️ TypeScript (Frontend), 🗂️ SQL.

  • Frameworks and Libraries ⚙️:
  • Platforms `☁️:
    • Gemini API: Uses Gemini 2.5 Flash for document parsing (PDF/image to structured transaction data) and forensic vision analysis.
    • Hugging Face: Hosts the GPT-OSS-120B inference endpoint used by both the Rust backend (LLM explanations, sanctions screening) and the Python ML agent pipeline.
    • Vercel: Frontend hosting and CI/CD.
    • Render: Backend API and ML microservice hosting and CI/CD.
  • Tools🪛: PostgreSQL (Primary database), Lucide React (Icons)

Product Summary 🎯

yosemite 🏔️ is an intelligent compliance and fraud detection dashboard designed for the modern business owner. It transforms raw financial data and documents into actionable insights through a sophisticated Intelligence Pipeline Flow.

Key Features 🔍

Concurrent Hybrid Analysis 🧠: When a document is uploaded, Yosemite triggers two parallel processes; AI-driven extraction parses unstructured text into structured transaction data, while Gemini Vision performs forensic analysis detecting visual tampering, font inconsistencies, and missing compliance fields that traditional scanners miss.

Multi-Agent Fraud Orchestration 🤖: 7 specialist AI agents: anomaly detection (Isolation Forest), Benford's Law analysis, duplicate invoice detection, Graph Convolutional Networks, BiLSTM sequence modeling, behavioral velocity analysis, and document vision. Running in parallel via Railtracks multi-agent orchestration framework. A coordinator agent synthesizes all findings into a single structured risk report with deterministic fallback guarantees, ensuring consistent results even if the LLM is unavailable.

Intelligent Triage 🚦: A three-tier scoring pipeline: transactions below 20 clear instantly, above 70 flag immediately, zero AI overhead. Only ambiguous cases (20–70) escalate to a GPT-class model with full transaction context. Powered by a Rust/Axum backend for rule scoring and a Python/FastAPI ML sidecar, our decoupled microservices delivering sub-1 second end-to-end latency across the full pipeline.

Explainable AI 💬: Converts multi-signal fraud findings into plain business language, summarizing why a transaction is suspicious and what action to take, producing executive-ready briefings without the noise.

Global Risk Mapping 🌍: Real-time sanctions screening against 1M+ entities via OpenSanctions API with fuzzy entity matching, compound geopolitical risk scoring, and AI-generated action recommendations for vendors in conflict zones or high-risk jurisdictions.

User Experience ⚯: A premium, responsive command center with high-level dashboards and deep-dive transaction forensics, inspired by Palantir.

Built With

Share this project:

Updates