Product Requirements Document (PRD)
Last updated: 2026-07-29
1. Product Overview
1.1 Vision
An autonomous, multi-agent patent idea generation and validation system that discovers patentable ideas from a knowledge base, processes them through rigorous validation workflows, and produces submission-ready patent packets — all powered by the LangChain DeepAgents runtime.
1.2 Goals
| Goal | Description |
| G1 | Automate patent idea discovery from knowledge base content |
| G2 | Provide rigorous multi-stage validation through 18 workflow states |
| G3 | Enable human-in-the-loop approval for critical review stages |
| G4 | Produce submission-ready patent disclosure documents |
| G5 | Maintain full audit trail with provenance metadata |
1.3 Non-Goals
| Non-Goal | Rationale |
| External patent API integration | Deferred; uses LLM knowledge + curated KB |
| Sandboxed code execution | Security risk; explicitly deferred |
| Database migration from YAML | Not warranted until scale requires it |
| Mobile app | Desktop-first web application |
2. User Stories
2.1 Patent Analyst
US-01: As a patent analyst, I want to submit a technical signal
so that the system can generate patentable ideas from it.
US-02: As a patent analyst, I want to see the workflow progress
of each idea in real time so that I can monitor processing.
US-03: As a patent analyst, I want to review agent reasoning and
tool calls so that I can trust the generated output.
US-04: As a patent analyst, I want to add comments to ideas so that
I can provide feedback during the workflow.
2.2 Manager/Reviewer
US-05: As a manager, I want to review and approve/reject ideas at
key workflow stages so that I can control quality.
US-06: As a manager, I want to see a review packet with scores and
evidence so that I can make informed decisions.
US-07: As a manager, I want to see pending approval requests so that
I don't block the workflow.
2.3 IP Counsel
US-08: As IP counsel, I want to validate patentability before
submission so that only viable patents are filed.
US-09: As IP counsel, I want to see prior art analysis and novelty
assessment so that I can evaluate filing strategy.
2.4 System Administrator
US-10: As an admin, I want to configure scoring weights and gate
checklists so that the system adapts to business needs.
US-11: As an admin, I want to see system statistics and analytics
so that I can monitor performance.
3. Functional Requirements
3.1 Idea Management
| ID | Requirement | Priority | Status |
| FR-01 | Create idea from manual signal text | P0 | Implemented |
| FR-02 | Autonomous idea seeding from KB | P0 | Implemented |
| FR-03 | List ideas with filters (phase, state, score) | P0 | Implemented |
| FR-04 | View full idea detail with all artifacts | P0 | Implemented |
| FR-05 | Update idea fields | P1 | Implemented |
| FR-06 | Delete idea (with approval interrupt) | P1 | Implemented |
| FR-07 | Archive idea (with approval interrupt) | P1 | Implemented |
| FR-08 | Duplicate idea detection | P2 | Implemented |
3.2 Workflow Engine
| ID | Requirement | Priority | Status |
| FR-09 | 18-state workflow machine | P0 | Implemented |
| FR-10 | Gate checklist validation per transition | P0 | Implemented |
| FR-11 | State transition with lifecycle hooks | P0 | Implemented |
| FR-12 | Advance to next state | P0 | Implemented |
| FR-13 | Advance to specific state | P1 | Implemented |
| FR-14 | Scheduler-driven autonomous cycles | P1 | Implemented |
| FR-15 | Full pipeline execution | P1 | Implemented |
3.3 Scoring
| ID | Requirement | Priority | Status |
| FR-16 | 7-criterion weighted scoring | P0 | Implemented |
| FR-17 | LLM-powered scoring | P0 | Implemented |
| FR-18 | Score history tracking | P0 | Implemented |
| FR-19 | Composite score with strength rating | P0 | Implemented |
| FR-20 | Filing threshold validation | P1 | Implemented |
3.4 Agent Runtime
| ID | Requirement | Priority | Status |
| FR-21 | DeepAgents runtime with middleware stack | P0 | Implemented |
| FR-22 | Subagent definitions from workflow roles | P0 | Implemented |
| FR-23 | Filesystem backend with permissions | P0 | Implemented |
| FR-24 | Skills and memory middleware | P0 | Implemented |
| FR-25 | HITL interrupt configuration | P0 | Implemented |
| FR-26 | Typed transcript event model | P0 | Implemented |
| FR-27 | Runtime event streaming | P0 | Implemented |
3.5 Human-in-the-Loop
| ID | Requirement | Priority | Status |
| FR-28 | Manager review interrupt | P0 | Implemented |
| FR-29 | IP review interrupt | P0 | Implemented |
| FR-30 | Counsel validation interrupt | P0 | Implemented |
| FR-31 | Delete/archive approval interrupt | P1 | Implemented |
| FR-32 | Approval/rejection endpoints | P0 | Implemented |
| FR-33 | Durable pending interrupt persistence | P0 | Implemented |
| FR-34 | Review analytics | P2 | Implemented |
3.6 Frontend
| ID | Requirement | Priority | Status |
| FR-35 | Dashboard with stats and idea list | P0 | Implemented |
| FR-36 | Idea detail view with tabs | P0 | Implemented |
| FR-37 | Live SSE updates | P0 | Implemented |
| FR-38 | Transcript-driven chat sidebar | P0 | Implemented |
| FR-39 | Agent todo/progress panel | P1 | Implemented |
| FR-40 | Subagent activity cards | P1 | Implemented |
| FR-41 | Tool call inspection | P1 | Implemented |
| FR-42 | Interrupt approval UI | P1 | Implemented |
| FR-43 | Artifact diff viewer | P2 | Implemented |
| FR-44 | Knowledge base browser | P1 | Implemented |
| FR-45 | Siemens controls page | P2 | Implemented |
3.7 Observability
| ID | Requirement | Priority | Status |
| FR-46 | LangSmith tracing | P1 | Implemented |
| FR-47 | Review analytics with reviewer identity | P2 | Implemented |
| FR-48 | Audit log via transcript events | P0 | Implemented |
4. Acceptance Criteria
4.1 Quality Gates
| Gate | Criteria |
| Tests | All existing tests must pass (pytest backend/tests) |
| No Fabrication | No silent fallback to fabricated agent output |
| Provenance | Every artifact carries provenance metadata |
| Error States | Failures are explicit (retry, pause, error) not hidden |
| Frontend Build | npm run build must pass with 0 errors |
| Metric | Target |
| API response time (non-streaming) | < 500ms |
| SSE event latency | < 200ms |
| State transition time | < 2s |
| Scoring time | < 5s |
| Full pipeline (3 ideas) | < 5 min |
5. Release Criteria
5.1 Phase Completion
| Phase | Criteria | Status |
| P0: Documentation + Scaffolding | Audit complete, docs written | Completed |
| P1: Backend Cleanup | Files split, routes modularized | Completed |
| P2: DeepAgents Runtime | Runtime factory, middleware, subagents | Completed |
| P3: Skills + Memory | Skills directory, memory files | Completed |
| P4: HITL Approvals | Checkpointer, interrupt endpoints | Completed |
| P5: Frontend Streaming | SSE adapter, transcript UI | Completed |
| P6: Artifact Quality | Versioning, diffs, provenance | Completed |
| P7: Advanced Capabilities | Multimodal, LangSmith, analytics | Completed |