Forseen

AI-powered compliance intelligence — anticipate regulatory change before it hits.

Forseen transforms a company profile into actionable regulatory insights. Complete a guided onboarding wizard, enter a compliance topic like “health data privacy,” and get live risk predictions, structured compliance reports, and interactive visualizations — all grounded in real regulatory signals from nine federal, state, and third-party sources. A RAG-powered chat interface lets you ask follow-up questions against your analysis and the retrieved documents.

---

Features

• Guided company setup — Multi-step profile wizard capturing industry, legal structure, data practices, operating states, and compliance posture with validation and persisted sessions
• Topic decomposition — Broad compliance questions are broken into specific regulatory sub-topics by K2 for targeted analysis
• Parallel predictions — K2 generates scored risk predictions (6-month, 12-month, 24-month probabilities) for each sub-topic concurrently
• Structured reports — Hermes produces plain-English compliance reports with executive summaries, priority actions, and section breakdowns
• Interactive dashboard — Risk radar, signal network graph, probability timelines, prediction cards, and drill-down modals
• Vector-backed retrieval — MongoDB Atlas vector search over signal embeddings for semantic matching
• RAG chat — Conversational follow-ups grounded in your analysis context and retrieved regulatory documents
• Nine regulatory sources — Federal Register, Congress.gov, FTC, FDA, CMS, ONC, NIST, CISA, state legislatures, Reddit, Google News, and NewsAPI
• Session persistence — Company profiles and analysis results persist across browser sessions via localStorage
• One-command local dev — Single script spins up all services including auto-starting MongoDB via Docker

---

Architecture

frontend/          React 19 + TypeScript + Vite (port 5175)
backend/           FastAPI orchestration API (port 8000)
model_servers/k2/  K2 Think v2 — topic decomposition & predictions (port 8001)
model_servers/hermes/  Hermes 4.3 — report & narrative generation (port 8002)
data_ingestion/    Signal pipelines: ingestors, embeddings, dedup, entity extraction, scoring
scripts/           Local dev scripts (run, stop, inspect)

---

Analysis flow

1. Profile — Complete the setup wizard (identity, scale, data practices, operating states, certifications)
2. Topic input — Enter a regulatory topic and jurisdiction
3. Signal fetch — Backend queries MongoDB with vector search for relevant regulatory signals
4. Decompose — K2 breaks the topic into 3-5 specific sub-topics
5. Predict — K2 runs parallel predictions across sub-topics with probability scores, reasoning, and counterfactors
6. Report — Hermes generates a structured compliance report from all predictions
7. Explore — Dashboard visualizations, drill-downs, and RAG chat for deeper questions

---

Built With

AI & Model Servers

• K2 Think v2 — Regulatory prediction and topic decomposition (OpenAI-compatible API)
• Hermes 4.3-36B — Structured report and narrative generation (OpenAI-compatible API)

Backend

• FastAPI — Orchestration API (/analyze, /predictions, /reports, /chat)
• Python 3.11+, Uvicorn, Pydantic 2.0
• Motor (async MongoDB driver) + MongoDB Atlas vector search
• httpx — Async HTTP for model server communication

Frontend

• React 19 + TypeScript 5.9 — Setup wizard, analysis UI, dashboard
• Vite 8 — Dev server and production builds
• Tailwind CSS v4 — Theming and layout
• Radix UI — Accessible component primitives
• Framer Motion — Animations and transitions
• Recharts — Data visualizations (radar, timelines, gauges)
• Sonner — Toast notifications

Data Ingestion

Nine ingestors feeding MongoDB with regulatory signals:

Source	Description
Federal Register	Rules, notices, proposed regulations
Congress.gov	Bills and legislation
FTC	Enforcement actions and guidance
FDA, CMS, ONC, NIST, CISA	Federal agency regulatory updates
State legislatures	All 50 states via LegiScan
Reddit	Compliance and regulatory subreddit discussions
Google News	Google News alerts
NewsAPI	News articles on regulatory topics

Pipeline stages: deduplication, entity extraction, summary enrichment, relevance scoring, signal velocity computation, and vector embedding generation.

---

Quick Start

Local (one script)

git clone https://github.com/<your-org>/YHack-2026.git
cd YHack-2026
./scripts/run-local.sh

This creates a Python venv, installs dependencies, and starts all four services. If Docker is available and MongoDB isn't running, it auto-starts a mongo:7 container.

Service	URL
Frontend	http://127.0.0.1:5175
API docs	http://127.0.0.1:8000/docs
K2	http://127.0.0.1:8001
Hermes	http://127.0.0.1:8002

Logs and PID files live under .logs/. Stop everything with:

./scripts/stop-local.sh

Docker Compose

docker compose up --build

Starts K2, Hermes, and the backend. Frontend runs separately via cd frontend && npm run dev.

Manual setup

# Backend
python3 -m venv .venv && source .venv/bin/activate
pip install -r backend/requirements.txt

# Frontend
cd frontend && npm install && cd ..

Environment

Copy .env.example files and fill in your keys:

File	Variables
backend/.env	MONGO_URI, MONGO_DB, K2_BASE_URL, HERMES_BASE_URL
model_servers/k2/.env	K2_API_KEY, K2_BASE_URL, K2_MODEL
model_servers/hermes/.env	HERMES_API_KEY, HERMES_BASE_URL_MODEL, HERMES_MODEL
frontend/.env	VITE_API_BASE_URL (defaults to http://localhost:8000)

---

API

Full reference: backend/API.md

Endpoint	Method	Description
/health	GET	Health check
/analyze/	POST	Full pipeline: signals + decomposition + predictions + report
/predictions/	POST	Direct K2 prediction (no signal fetch)
/reports/	POST	Direct Hermes report from prediction IDs
/chat/	POST	RAG chat with signal and prediction context

Example:

curl -s -X POST http://127.0.0.1:8000/analyze/ \
  -H “Content-Type: application/json” \
  -d '{
    “topic”: “health data privacy”,
    “jurisdiction”: “CA”,
    “company”: {
      “name”: “Acme Health”,
      “legal_structure”: “C-Corp”,
      “industry”: “Healthcare SaaS”,
      “size”: 45,
      “location”: “San Francisco, CA”,
      “operating_states”: [“CA”, “NY”],
      “description”: “B2B clinical decision support”,
      “handles_pii”: true,
      “handles_phi”: true,
      “uses_ai_ml”: true,
      “b2b”: true,
      “certifications”: [“HIPAA”]
    }
  }'

---

Acknowledgments

Built at YHack 2026. Thanks to the organizers, mentors, and sponsors.

---

Future Roadmap

• International regulatory sources (EU, UK)
• Real-time monitoring and alerts for new regulations
• Historical trend analysis for regulatory pressure
• Collaborative features for compliance teams
• GRC platform integrations
• Domain-specific model fine-tuning
• Auth and multi-tenant company profiles
• CI pipeline (lint, typecheck, tests)

---