Bhairyam AI is a sovereign legal AI workspace for UK legal teams. It is built for matter-based research, document review, legal drafting, authority extraction, and source-grounded analysis over private PDF and Word files.
The architecture is intentionally self-hosted. The frontend runs on Next.js, ingestion and retrieval run through a FastAPI backend, source documents are stored locally by workspace, and semantic search uses Postgres with pgvector. Ollama powers local chat and embeddings so a firm can demonstrate private legal AI without depending on a generic SaaS chatbot for sensitive matter files.
What Bhairyam AI Does
Bhairyam AI lets a legal team create a matter, upload source documents, ingest PDF and Word files, extract legal citations, and ask questions against the retrieved evidence. The system focuses on UK legal research and matter-grounded answers rather than broad generic chatbot behaviour.
- Document ingestion: PDF and Word uploads are parsed, chunked, embedded, and stored for retrieval.
- Legal RAG: questions are answered from matter documents through pgvector retrieval and source-aware prompts.
- Authorities: cited cases, legislation, and legal references are extracted into a reviewable workspace.
- Matter views: documents, sources, authorities, history, issues, chronology, evidence, and audit views keep legal work organised.
- Drafts and Agreements: reusable skills generate neatly formatted DOCX outputs for agreement review and drafting workflows.
- Voice input: users can dictate questions instead of typing.
Why It Matters
Legal AI demos often fail when they look polished but cannot prove where an answer came from. Bhairyam AI is designed the other way around: documents, citations, authority lists, retrieved sources, and matter context are first-class parts of the product.
Production Architecture
The product is structured as a single repository with separate frontend and backend apps. Docker Compose starts the web app, FastAPI backend, Postgres pgvector database, and local storage volumes so the system can be demonstrated as a sovereign legal AI stack.
- Frontend: Next.js application with responsive chat, workspace, skills, and authentication screens.
- Backend: FastAPI services for document parsing, storage, ingestion, retrieval, and research synthesis.
- Database: Postgres with pgvector for embeddings, chunks, matters, authorities, users, audit events, and research history.
- Storage: local source folders instead of third-party object storage.
- Models: Ollama Qwen for generation and local embedding models for retrieval.
Designed For UK Legal Work
Bhairyam AI is focused on UK legal use cases: solicitor document review, case and legislation extraction, employment and commercial agreement review, matter chronology, issue spotting, and source-grounded legal research memos. It is not positioned as a lawyer replacement; it is a private matter intelligence workspace that helps legal professionals work through documents faster.