Turn messy customer feedback into an evidence-backed product roadmap.
SignalForge is a portfolio-grade product management case study built by an AI/ML student to demonstrate PM thinking, product strategy, analytics, prioritization, experimentation, and technical fluency.
Feedback → Evidence → Opportunities → Prioritization → Roadmap → PRD → Experiment → Decision log
The prototype is intentionally transparent: instead of hiding a recommendation behind a black-box AI score, it exposes the assumptions a PM can challenge.
Real product teams receive feedback from support tickets, app reviews, interviews, sales calls, and community posts. The hard part is not collecting feedback — it is turning noisy anecdotes into a decision the team can defend.
SignalForge addresses four common failure modes:
- Loud-user bias: the most recent request wins.
- Feature framing: teams prioritize requested solutions instead of underlying problems.
- False precision: a single score hides its assumptions.
- Decision amnesia: teams forget why a roadmap choice was made.
See evidence coverage, decision velocity, opportunity health, and the current recommendation at a glance.
Inspect customer problem clusters and drill into their source, volume, severity, confidence, and evidence tags. Add a new signal directly in the interface.
Use opportunity scoring and RICE-style reasoning. The assumptions are visible so reviewers can disagree with the model rather than blindly trusting it.
Move opportunities between Now / Next / Later to simulate a prioritization conversation and make tradeoffs visible.
Translate the selected problem into a concise PRD brief, primary metric, guardrails, instrumentation plan, and success criteria.
Preserve the reasoning behind roadmap choices: evidence, framework, confidence, and status.
If PMs can move from anecdotal feedback to a transparent prioritization decision in minutes, roadmap conversations become less opinion-driven and more evidence-driven.
% of roadmap decisions linked to ≥3 independent evidence items.
Initial product targets:
| Metric | Target |
|---|---|
| Evidence-linked decisions | 70%+ |
| Raw feedback → ranked opportunity | <10 min |
| Reviewer agreement with ranking | 80%+ |
Transparency over black-box automation — the score is explainable.
Problems over feature requests — customer language is synthesized into underlying friction.
Metrics before shipping — every meaningful roadmap bet gets a measurable hypothesis and guardrails.
Quality of decision over breadth of integrations — v1 avoids Jira/Linear/auth complexity until the core workflow proves useful.
- Vanilla HTML / CSS / JavaScript
- No build step
- Responsive SaaS-style interface
- Client-side interactive state for the portfolio prototype
- Render Static Site deployment
- Synthetic, traceable sample data
The intelligence layer is intentionally lightweight and transparent. The portfolio story is about product reasoning, not pretending a mock interface is production AI.
SignalForge/
├── app/
│ └── index.html # Interactive product prototype
├── docs/
│ ├── PRD.md # Product requirements
│ ├── ROADMAP.md # Prioritization and sequencing
│ ├── EXPERIMENT.md # Validation plan
│ └── CASE_STUDY.md # PM case study
├── render.yaml # Deployment configuration
└── README.md
The next validation step is a lightweight controlled usability test:
- 5–8 target users
- Control: spreadsheet / notes workflow
- Treatment: SignalForge
- Primary metric: median time to a ranked top-three list
- Guardrails: reviewer confidence, evidence coverage, contradictory signals ignored
- Success threshold: ≥30% reduction in median decision time while maintaining or improving confidence
The goal is to learn whether structured evidence actually changes decision quality — not merely whether users think the interface looks good.
NOW
- Interactive signal inbox
- Transparent opportunity scoring
- RICE reasoning
- Now / Next / Later roadmap
- Decision traceability
- PRD + experiment workflow
NEXT
- 5–8 user interviews
- Usability test against a spreadsheet control
- Capture actual decision time and confidence
- Tune weights from observed behavior
LATER
- CSV/API ingestion
- Jira / Linear integrations
- LLM-assisted synthesis with citations
- Team workspaces and permissions
- Historical outcome learning
https://signalforge-52lf.onrender.com
This project is a PM portfolio artifact, not a claim of production readiness. Sample metrics and feedback are synthetic and are clearly labeled so the reviewer can focus on the quality of the product decisions.
Built by Soniya — AI/ML student exploring product management at the intersection of user problems, data, experimentation, and technology.