Installed Novus

🚀 ShipSense — AI-Native Product Intelligence Studio

The world's most ambitious Product Command Center: from fuzzy idea to shipped, instrumented product — in a single window.

💡 Inspiration

It was a Wednesday afternoon when our lead PM had 11 browser tabs open — Notion for the PRD, Jira for the backlog, Miro for personas, Typeform for survey results, Amplitude for analytics, ChatGPT for brainstorming, G2 for competitor research, Dovetail for feedback clustering, and three Slack threads demanding answers she didn't have yet.

Sound familiar?

The Context Switching Tax is the silent killer of great product work. Research shows knowledge workers lose 23 minutes of deep focus every time they switch contexts. For a PM juggling 10+ tools daily, that's hours of compounding cognitive debt — hours that should be spent talking to customers, not copy-pasting between tabs.

"I don't need another AI tool. I need one AI tool that makes all the other ones unnecessary." — Every PM we interviewed

ShipSense was born from that frustration. We set out to build the world's most ambitious AI-native Product Intelligence Studio — a single window that takes a PM from fuzzy idea all the way to a shipped, instrumented product. No tabs. No context loss. No excuses.

⚡ What It Does

ShipSense is a full-lifecycle Product Command Center with 6 deeply interconnected AI modules and 30+ features. Every module shares a single reactive brain — data flows automatically, sessions persist across refreshes, and the AI actively coaches you on what to do next.

🗺️ The Complete Product Intelligence Suite

#	🧩 Module	🎯 One-Line Value	✨ Standout Features
1	🔥 IdeaForge	Validate before you build	Ship Score™ (0–100 composite), TAM/SAM/SOM Market Pulse, Devil's Advocate risk analysis, Competitor Radar with threat levels, Pain Point Validator with real user quotes
2	📄 PRD Architect	From idea to engineering brief in 60 seconds	Full PRD generator, User Stories with Gherkin AC, PRD Health Score with section-by-section diagnosis, Version Timeline, one-click PDF export
3	👤 Persona Lab	Know your user before you build for them	AI Persona generator, Jobs-To-Be-Done Canvas, 6-quadrant Empathy Map, Behavioral Prediction engine with probability scores, Persona Conflict Analyzer
4	🎯 PriorityMatrix	Score with data, not gut feel	Dynamic RICE Scorer, MoSCoW auto-classifier, Dependency Web mapper, Monte Carlo Sprint Simulator with team velocity modelling
5	💬 FeedbackFusion	Turn 500 support tickets into 5 actionable themes	AI Feedback Clustering, Sentiment Trajectory chart, Feature Request Ranker, NPS Verbatim decoder, Competitive Intel digester
6	🛸 LaunchControl	Ship with total confidence	GTM Canvas, Changelog AI (major/minor/patch), Success Metrics (leading + lagging + guardrails), Stakeholder Brief generator, 40-item Ship Readiness Checklist

🧠 The Living Product Brain — Cross-Module Connectivity

Unlike any other tool, ShipSense modules are not siloed. They share a single reactive state layer that creates a seamless, living product workflow:

FeedbackFusion ──[Push to PriorityMatrix 🎯]──▶ PriorityMatrix
                                                       │
                                               [Draft PRD ⚡]
                                                       │
                                                       ▼
                                               PRD Architect
                                                       │
                                           [Navigate to LaunchControl →]
                                                       │
                                                       ▼
                                               LaunchControl

Real example in 4 clicks:

💬 Paste 50 customer support tickets into FeedbackFusion → AI clusters them into themes
🎯 Click "Push to PriorityMatrix" on the #1 requested feature → instant RICE score appears
⚡ Run Sprint Simulation → click "Draft PRD" on the winning feature → full PRD streams in real-time
🛸 Navigate to LaunchControl → your validated idea is pre-populated → generate your GTM canvas

Total time: under 3 minutes. In any other tool stack, that workflow takes days.

🤖 Hybrid AI Engine — The Best of Both Worlds

Mode	How It Works	When It Runs
⚡ Instant Mode (default)	Proprietary deterministic hashing engine — generates rich, realistic, contextually-accurate data in `< 5ms`	Always — no API key needed
🧠 Hybrid AI Mode (BYOK)	Real OpenAI `gpt-4o-mini` completions via user's own API key — fully dynamic, edge-case-proof responses	When user adds key via ⚙️ Settings

Judges test with edge-case inputs like "SaaS for dog walking on Mars." Our deterministic engine handles it gracefully. With an OpenAI key, it handles it brilliantly.

🛠️ How We Built It

We made one bold architectural decision: zero framework, zero build step, zero compromise.

🏗️ Layer	🔧 Technology	💭 Reasoning
Core	Vanilla HTML5 / CSS3 / ES Modules	No React, no bundler — loads in milliseconds, runs everywhere, judges can inspect source directly
State	Custom reactive pub/sub `State` manager	Session-persistent via `localStorage` with safe deep-merge — survives hard refreshes, never corrupts
AI Engine	Hybrid BYOK Architecture	Instant deterministic results by default; real LLM completions when user brings their OpenAI key
Telemetry	Novus.ai (Pendo SDK)	Live production SDK firing `module_activity`, `score_increased`, and `prd_pdf_downloaded` events to a real dashboard
PDF Export	`html2pdf.js` CDN	Renders fully-styled PRD HTML into a polished, professional PDF — PMs live in PDFs, not Markdown
Design System	Vanilla CSS with CSS custom properties	Glassmorphism, animated SVG score rings, particle canvas, micro-animations — all hand-crafted
Deployment	Vercel (static hosting)	Zero-config, global CDN, sub-100ms TTFB worldwide

📐 Architecture Diagram

┌─────────────────────────────────────────────────────────────┐
│                        index.html                           │
│              (1 file · 1,100+ lines · zero deps)            │
└───────────────────────┬─────────────────────────────────────┘
                        │ ES Module imports
              ┌─────────▼──────────────────┐
              │          js/app.js          │
              │  Routing · Particles · Dashboard  │
              │  Lifecycle AI Coaching Engine     │
              └──┬──────────────────────┬───┘
                 │                      │
      ┌──────────▼──────────┐  ┌────────▼────────────────┐
      │   js/ai-engine.js   │  │   js/utils/state.js     │
      │   Hybrid AI BYOK    │  │   Reactive pub/sub      │
      │   30+ generators    │  │   localStorage persist  │
      │   OpenAI intercept  │  │   Novus.ai telemetry    │
      └──────────┬──────────┘  └────────┬────────────────┘
                 │                      │
      ┌──────────▼──────────────────────▼─────────────────┐
      │                  6 Module Files                    │
      │  ideaforge · prd-architect · persona-lab           │
      │  priority-matrix · feedback-fusion · launch-control│
      └────────────────────────────────────────────────────┘

🚧 Challenges We Ran Into

Every great product has battle scars. Here are ours — and exactly how we won each fight:

⚔️ Challenge	🎯 Root Cause	🏆 Solution
The Blank Slate Problem	Judges won't type 50 lines of feedback to see a clustering demo	Built glowing `btn--glow` one-click sample loaders that populate every module with rich, realistic, interconnected data
The Async Crash Bug	Upgrading `generatePRD()` to `async` for BYOK broke all 6 module call sites silently	Audited every call site and added precise `await` — found and fixed 2 critical missing awaits
Cross-Module DOM Events	"Push to PriorityMatrix" buttons live inside AI-generated HTML — standard listeners dead on arrival	Implemented `document.body` event delegation using `e.target.classList.contains()` for reliable dynamic element handling
State Corruption on Refresh	Old `localStorage` sessions had missing schema keys, causing silent `undefined` crashes	Implemented `safeLoadState()` — a deep-merge strategy that fills missing keys from defaults and resets cleanly on parse errors
SVG Ring Animation Bug	`stroke-dashoffset` animation failed because `stroke-dasharray` wasn't set inline first	Fixed `animateRing()` to set `strokeDasharray` inline before the CSS transition fires
`effortCls` Class Mismatch	`scoreFeature()` returned CSS class names that didn't match actual CSS definitions	Found and fixed the `effort-xs` / `eff-xs` naming mismatch — RICE badges now render with correct color coding

🏆 Accomplishments We're Proud Of

🥇 Proving it's a real product

🔴 Live Novus.ai Telemetry — Real production Pendo SDK fires custom events to an active dashboard. Not a console.log. A live funnel.
💾 Crash-Proof Session Persistence — A PM's entire workflow (PRDs, personas, scored features, checklist state) survives a hard browser refresh with our safe deep-merge state architecture.
📄 PDF PRD Export — Click Download in PRD Architect and receive a polished, print-ready PDF. PMs send PDFs to engineering. We export PDFs.
🤖 Real LLM Integration — BYOK mode sends live requests to gpt-4o-mini. The fallback is instant. The upgrade is transformative.

🎨 Proving the craft

⚡ Zero-Latency UX — The deterministic AI engine returns results in < 5ms. There are no loading spinners in the critical path.
🧠 AI Lifecycle Coaching — checkLifecycleHealth() monitors PM progress and proactively intervenes: "Your idea is validated but you have no Persona. Build one now." This is what a great co-pilot does.
🌐 30+ Features, 0 External UI Libraries — Every animation, ring, card, and micro-interaction is hand-crafted CSS. No Tailwind. No Material UI. Just craft.

📚 What We Learned

"The final 2% of polish is where first place is decided."

💡 Lesson	📖 The Insight
Instrumentation proves product thinking	Adding Novus.ai with custom event schemas showed we think in funnels and retention — not just features. Judges notice this.
Empty states are your first impression	Every judge opened PriorityMatrix and saw an empty list. Adding a glowing "Load Sample Framework" button removed 100% of that friction.
Async is a contract across the whole codebase	One `async` keyword in `ai-engine.js` created a ripple of required `await` calls across 6 modules. Missing one crashes silently.
State persistence is the "realness" litmus test	Any real PM tool survives a page refresh. We made ours survive it — and added safe error recovery when the schema changes between versions.
Cross-module data flow is the moat	Any PM can copy our UI. Copying the seamless "Push to PriorityMatrix → Draft PRD → Launch" workflow requires rebuilding the entire reactive state architecture. That's a real moat.

🚀 What's Next for ShipSense

We're not stopping at the hackathon. Here is the ShipSense 2026–2027 Roadmap:

🗓️ Timeline	🔮 Feature	📋 What It Unlocks
Q3 2026	🔗 Jira / Linear Two-Way Sync	PRD Acceptance Criteria push directly to tickets; ticket status pulls back into PRD health score
Q3 2026	🎨 Figma Plugin	Embed live UI mockups inside PRD Architect; design and spec live side-by-side
Q4 2026	👥 Multiplayer Mode	Real-time collaborative RICE scoring — PM, Engineering, and Design score together, live
Q4 2026	🎙️ Voice Input	Dictate ideas, feedback, and feature names via microphone → instant AI processing
Q1 2027	🤖 Slack / Teams Bot	`/shipsense score [feature]` directly from any Slack channel — zero context switch
Q1 2027	🧬 Custom Persona Training	Upload your own user interview transcripts → ShipSense generates hyper-specific, proprietary personas
Q2 2027	📊 Analytics Integration	Connect Amplitude or Mixpanel → feature usage data auto-feeds RICE scores with real confidence intervals