BridgeAI
AI-native requirements engineering platform
Traditional requirements gathering between clients and business analysts is slow, inconsistent, and error-prone: days-to-weeks of back-and-forth, missed details, no standardization across projects, and no real audit trail of how a requirement evolved.
A client describes what they want to build in a normal chat interface. A multi-agent system built on Claude 3.5 Sonnet and Claude 3 Haiku detects ambiguity, asks targeted clarifying questions, extracts structured requirements, and assembles them live into a formatted CRS document streamed to the browser over WebSocket as it's generated. A business analyst then reviews, comments inline, and approves or rejects the result — every step versioned and audited.
The product itself — the agent pipeline, the app, the full feature set — is built and functional locally. What remains is purely the production deployment: a pre-launch hardening pass (dead background workers now actually run, CORS fixed, secrets rotated, internal Docker ports no longer exposed, TLS wiring written) is done in the working tree but blocked on provisioning a VPS.
Engineering highlights
- Clarification agent detects ambiguity and asks targeted follow-up questions before generating anything, classifying intent (requirement vs. greeting vs. question).
- Template filler agent extracts and maps conversation content into a structured CRS across 15+ sections, with per-field source tracking and SHALL/SHOULD/MUST/MAY/WILL vocabulary enforcement.
- Memory agent stores requirements as embeddings in ChromaDB, scoped per project, for semantically relevant recall across a conversation.
- Multi-standard CRS templates — BABOK, IEEE 830, ISO/IEC/IEEE 29148, and Agile user-story formats — selectable per project, with full version history and an audit trail.
- Dual-role UX for clients and business analysts, with team workspaces, RBAC (Client / BA / Owner / Admin / Member / Viewer), and export to PDF, Markdown, and CSV.