Capstone project for Applied Agentic AI for PMs/TPMs. PRD Genie ingests meeting transcripts, product briefs, and stakeholder notes, and automatically produces structured requirements, a Product Requirements Document, Epics & User Stories, and a Gap Analysis — while explicitly refusing to invent requirements that aren't grounded in the source material.
Sequential Pipeline (see docs/architecture-diagram.png and the Q3 section of
docs/PRD_Genie_Q1_Q2_Q3_Q4.docx for full rationale):
Transcript / Notes / Brief
↓
Requirement Extractor Agent
↓
PRD Generator Agent
↓
Epic & User Story Generator
↓
Gap Analysis Agent
↓
Final Documentation Agent
- Import
workflow/PRD_Genie_Workflow.jsoninto n8n. - The export does not include API credentials — add your own OpenAI credential in n8n and attach it to each of the 5 Chat Model nodes (one per agent).
- Open the "Transcript Input" node, paste a transcript, click Execute Workflow.
All 12 cases from the baseline dataset (eval_prdgenie_inputs.txt) were run
against the problem statement's "Expected Output Must Contain" criteria.
12/12 pass after prompt-level fixes — see evaluation/baseline_test_results.md
for the full before/after results table, including every bug found and fixed
during testing (contradiction handling, verbatim acceptance criteria,
invented benefits, competing-proposal merging, and empty-input handling).
Observability: traced via n8n's native Execution History (see
screenshots/ for example execution traces) rather than a dedicated
LLM-ops tool, due to project time constraints — see Q3 for the full
justification.
See demo/demo_video_link.md for the 5-minute walkthrough video.
docs/PRD_Genie_Q1_Q2_Q3_Q4.docx— Q1 Ideation, Q2 Program Charter, Q3 Design/Build/Cost & Evaluation Strategy, Q4 Reflectiondocs/PRD_Genie_Slide_Deck.pptx— presentation summarydocs/architecture-diagram.png/.svg— pipeline architecture diagramevaluation/baseline_test_results.md— full 12-case baseline resultsworkflow/PRD_Genie_Workflow.json— exported n8n workflow