PROJECT 03 · ROADMAPPING / DECISION SUPPORT
A capacity-aware planning tool that turns a messy backlog into an explainable roadmap, then helps product teams explore risks, dependencies, delivery scenarios, and follow-up questions with AI.
THE PROBLEM
Product teams often maintain backlogs with inconsistent scoring, hidden dependencies, and little connection to delivery capacity. The result can look ordered while still leaving basic questions unanswered: what fits, what slips, and why?
HOW THE MVP WORKS
The MVP connects prioritization, capacity planning, dependency logic, visualization, and AI assistance in one flow. Users can start manually, import a CSV, explore editable sample data, and export the completed plan.
Choose a prioritization method, team size, delivery velocity, and planning window to define the constraints behind the roadmap.
Capture initiatives, outcomes, effort, confidence, dependencies, and framework-specific inputs manually, through CSV import, or with editable sample data.
Scores initiatives with Simple, RICE, or WSJF, then builds a roadmap around available capacity and dependencies. Users can view it as a board or timeline and export it as CSV.
Answers questions about timing, risk, capacity changes, dependencies, and outcomes using the generated roadmap as its bounded source of truth.
ROADMAP INSIGHTS
Once a roadmap is generated, deterministic insight cards highlight how effort is distributed across outcomes, where outcome gaps exist, and how dependencies may affect delivery. This gives product teams an immediate decision layer while providing grounded context for deeper AI follow-up questions.
PRODUCT DECISIONS
Each framework exposes its inputs and formula. Product judgment remains visible, and users can edit assumptions before regenerating the roadmap.
The roadmap is built from team size, velocity, effort, and dependencies — not only from priority order.
Deterministic calculations are reused when available. The assistant receives structured roadmap context, distinguishes facts from interpretation, and says when the available data is insufficient.
GROUNDED AI IN PRACTICE
The assistant receives the generated roadmap and recent conversation as structured context. Scores, capacity, timing, and dependencies are calculated deterministically in the application, while a cost-efficient Gemini model explains those facts and connects relevant signals. It is instructed to distinguish facts from interpretation and acknowledge when the roadmap lacks sufficient information, reducing the risk of hallucinated answers. A per-user daily question limit also keeps operating costs predictable and discourages abuse.
OUTCOME & LEARNING
The MVP supports the complete journey from backlog entry to prioritized plan, delivery-confidence signals, export, and follow-up analysis. Responsive board and timeline views make the same planning model usable across desktop and mobile.
The strongest roadmaps begin with explicit priorities and intended outcomes. AI adds value when guardrails keep it grounded in that plan — helping teams explore trade-offs without hallucinating roadmap facts.