Requirements Manager
Agentic requirements management
The Requirements Manager (rm.reqpool.com) is the suite's agentic requirements management platform: it covers the entire path from raw sources — documents, video, audio — to a confirmed, exportable specification, with AI in every process step. Operated on Azure with Entra ID SSO; the specification lives in a graph database (Neo4j), AI models run via Azure AI Foundry (GPT-5, Claude Opus/Sonnet), transcription via Azure AI Speech.
Modules & features
Structured specification as a tree
Modules group epics, epics group user stories. Every requirement has Markdown text, a requirement type (functional, non-functional, constraint, interface, data, quality), status, comments and a complete version history — every change creates a new version.
AI extraction from any source
Documents (Word, PDF), videos and audio are uploaded and analyzed agentically; audio/video is transcribed with speaker attribution. Every extracted requirement gets a source reference — with a timestamp for media; the module/epic hierarchy is built directly from the sources. Processing runs as background jobs with progress display and the option to cancel.
Quality module
Analysis of every requirement against eight criteria (unambiguous, complete, consistent, verifiable, atomic, necessary, feasible, traceable) plus document-level checks (completeness, freedom from redundancy and contradiction, uniform structure, prioritization). Quality scores per module and epic; weak formulations are highlighted in the text, and corrections can be adopted individually or rolled out agentically across the entire specification.
Engineer: getting ready to build
Analyzes the specification from an implementation perspective (programming, CI/CD, operations), identifies all open points and phrases them as questions to the product owner — each with a pre-filled AI assumption to simply confirm or overwrite. The answers automatically become new requirements in the specification.
Export/import with a branch model
Export to Excel, Word, ReqIF and JSON. Every export is a working branch: on re-import the platform compares base, main trunk and branch per requirement (three-way merge); the AI detects semantic conflicts and duplicates and proposes the resolution — clients and business units work offline in Excel/Word without versions drifting apart.
Baselines & coding intent
Freeze version states and have them confirmed — referenceable for contracts, estimates and acceptance, and direct input for ReqPOOL Signoff. On request the platform generates an agentic coding intent as Markdown — the bridge to AI-supported implementation.
MCP interface for AI agents
The backend provides an MCP endpoint (streamable HTTP, secured via Entra ID): external AI clients and coding agents can search, read and edit the specification with role-based permissions.
Governance & administration
SSO exclusively via Microsoft Entra ID, invitation by e-mail; roles system admin, admin/project manager, user and client. The system admin manages AI models centrally; per project a strong model (extraction, conflict resolution) and a fast model (classification, bulk work) are selected — including data zone labeling (EU/global).
Customer value
- Faster to specification: Workshop recordings, requirement documents and e-mails become a structured, source-linked specification in hours instead of weeks.
- Provable quality: Objective quality criteria and scores make maturity measurable; agentic improvement raises it systematically — less rework and fewer misunderstandings in implementation.
- End-to-end traceability: Source references down to the timestamp in a video, a version history for every requirement, frozen baselines — reliable for contracts, acceptance and audits.
- Collaboration without media breaks: Clients work in the platform with their own role or offline in Excel/Word; the AI-supported merge catches conflicts.
- Less project risk: Engineer uncovers all open points before implementation starts, instead of discovering them mid-sprint.
- AI-ready in both directions: Requirements are created with AI support and flow via coding intent and MCP directly into AI-supported development — the specification becomes the machine-readable basis of implementation.
- Enterprise-grade: Entra ID SSO, role model, central model management with an EU data zone option, operated on Azure.
Usage best practices
- Record workshops consistently — the platform works with original sources: transcription with speaker attribution and timestamped source references instead of memory protocols.
- Run the quality module before every gate; review agentic improvements with the business side before adopting them.
- Work through Engineer completely before implementation starts — have every AI assumption confirmed or overwritten by the product owner.
- Involve business units and clients via the client role or the Excel/Word branch; re-import with the AI merge instead of manual reconciliation.
- Have baselines confirmed before contracts, estimates and acceptance; review the coding intent together with architecture before it goes to implementation.