Magnai | UK Public Affairs

Magnai is a vertical AI SaaS platform that automates public affairs monitoring for UK Parliament activity, media coverage, and social media. Built for public affairs teams, it transforms hours of manual research into automated, AI-generated intelligence briefings.

- Backend: Django, Python, PostgreSQL (Supabase)

- AI/ML: OpenAI GPT-4, Claude Agent SDK, multi-agent orchestration

- Infrastructure: Docker, Railway, Django Q (async tasks), WebSockets

- APIs: UK Parliament API, Twitter API, Opoint Media API

Multi-agent AI system that monitors parliamentary debates, written questions, votes, bills, and media coverage. Generates structured briefings with source attribution and Hansard links.

- Control Agent: Orchestrates sub-agents (parliament, media, Twitter)

- Parliament Agent: Searches debates, questions, votes, bills via UK Parliament API

- Media Agent: Monitors news coverage via Opoint API

- Background Processing: Django Q task queue for scheduled reports

Autonomous research agent that identifies relevant MPs and Lords for any policy topic.

- Claude Agent SDK: Agentic research with MCP tools wrapping Parliament API

- Intelligent Search: Searches debates, written questions, EDMs, committees, bills

- Evidence-Based: Returns member IDs with rationale, confidence scores, and source citations

- Auto-Enrichment: Hydrates stakeholder profiles from Parliament Members API

- Real-time Progress: WebSocket updates during map generation

Track parliamentary bills through their lifecycle with AI-powered summaries and amendment analysis.

Team-based access control with company/team hierarchy. URL-based tenant isolation /<company_slug>/).

- LLM Tool Schema Design: Solved GPT-4 hallucination issues by removing optional params that trigger dummy value injection

- Claude Agent SDK Deployment: Fixed macOS ARM64 binary issues in Docker Linux containers via npm-based CLI installation

- Parliament API Integration: Full client for Members, Bills, Votes, Committees, Oral/Written Questions endpoints

- Race Condition Handling: Django Q tasks with proper locking for concurrent report generation

- Reduced manual parliamentary monitoring from hours to minutes

- Automated stakeholder identification with evidence-based rationale

- Real-time briefings with direct Hansard links and source attribution