Inspiration 💡
PMs lose hours to copy/paste glue across Jira/ADO, Confluence, email, and chats.
Strategy/KPIs rarely flow into stories—sprints drift from goals.
Migrations (e.g., Jira ↔ ADO) create tool chaos and duplicated work.
Lost in pages: decisions and context get buried across dozens of Confluence/Docs pages—ProDuh anchors a single canonical page per phase, auto-links related artifacts, and keeps My Notes attached so nothing gets lost.
We wanted an AI layer that removes busywork so PMs can focus on decisions and people.
🖥️ What it does
Multi-agent, LLM-powered command center for the PM lifecycle: Discovery → Strategy → Backlog → Jira Enhancements → GTM → Release → Operate.
App-agnostic: works with Jira or ADO (plus Confluence, MS Graph, ServiceNow, GitHub read).
Strategy → KPIs → Stories: turns goals into Jira/ADO-ready stories with AC, tasks, and QA checklist.
Jira Enhancements: RICE/WSJF hints, dependency detection, DoD checklist.
Docs & Notes: auto-create/update Confluence pages; My Notes for links and references.
Human-in-the-loop with optional autonomous runs for repeatable flows.
😎 How we built it
Multi-agent orchestration (Google ADK concepts + LLMs) with capped loops for determinism.
Prompt engineering (e.g., DevelopmentTasksPrompt) to guarantee Jira-ready formats.
Universal adapter layer for Jira/ADO parity; drivers for Confluence, MS Graph, ServiceNow, GitHub (read).
Mock analytics CSV to prove KPI→Operate feedback.
Node/Express backend, TypeScript/React frontend, JSON I/O for demo reliability.
🆘 Challenges we ran into
Normalizing Jira vs ADO schemas and status flows.
Balancing autonomy vs. oversight (keeping PM in control).
Enforcing consistent acceptance criteria and story quality.
Avoiding “LLM drift” → added guardrails and ≤3 loop cap.
Keeping the demo fast while integrating multiple systems.
🚀 Accomplishments that we're proud of
End-to-end flow from messy discovery notes to structured KPIs to Jira/ADO-ready stories.
Priority & dependency intelligence that actually changes planning conversations.
App-agnostic adapters so PMs can manage mixed stacks.
Tight 2-minute demo that shows real PM pain disappearing.
📕 What we learned
PMs don’t want another tool—they want a command center that talks to their tools.
Clear, enforced output schemas beat clever prompts alone.
Small QA micro-agents (requirements checks, DoD) deliver outsized value.
Strategy telemetry must be designed before shipping, not after.
📈 What’s next for ProDuh!
Deeper analytics loop: auto tracking plans, KPI binding, Operate insights.
Risk & compliance agent for regulated environments.
Roadmap coherence checks to catch drift early.
Org policy packs (naming, AC/DoD templates, GTM checklists).
Bi-directional Jira↔ADO sync with conflict resolution.
Enterprise hardening: SSO, SOC2 guardrails, PII redaction.
Built With
- adk
- agentic-architecture
- ai-agent
- gemini
- google-cloud
- llm
- openai
- postgresql
- prompt-engineering
- react
- sql
Log in or sign up for Devpost to join the conversation.