Tarazu

Decision intelligence for product teams. Weigh what matters.

Tarazu helps product teams prioritize ideas, compare tradeoffs, capture context, and generate explainable recommendations with structured frameworks like RICE — powered by Claude.

→ Live Demo · → Read the PRD

RICE-scored candidates on the left, the AI Decision Advisor's recommendation on the right. Try it live →

---

What It Does

Feature	Description
Normalized RICE Scoring	Slider-based input for each dimension on a 1–100 scale with real-time score calculation
Priority Matrix	Canvas-rendered Effort vs. Impact scatter plot with labeled quadrants
AI Strategy Advisor	One-click backlog analysis via Claude Opus — returns top priority, quick win, risk flag, sprint plan, and strategic insight
AI Score Suggestions	Per-candidate scoring via Claude Sonnet, grounded in product context and prior feedback
Persistent Storage	Features save across sessions via localStorage and cloud sync
Responsive Shell	Three-panel desktop layout (left rail / center canvas / right rail) collapses to a two-column tablet overlay and a bottom-tab mobile layout via matchMedia

A Closer Look

From a scored backlog to an AI recommendation in one click:

Tradeoff map	Candidate detail
Effort × Impact with QUICK WIN / STRATEGIC / FILL-IN / AVOID quadrants	Per-candidate RICE breakdown, formula, and metadata

Why I Built This

Prioritization often consumes hours per sprint-planning cycle, especially when teams rely on spreadsheets. Tarazu replaces that workflow with a purpose-built decision system that enforces RICE discipline, visualizes tradeoffs, and adds AI analysis that would otherwise require a senior PM or consultant.

It sits at the intersection of product management domain expertise and AI engineering — the exact skillset I bring to PM and technical leadership roles.

Tech Stack

Layer	Choice	Why
Frontend	React + Next.js	Component model, fast builds, file-based routing
Visualization	Canvas 2D API	No library dependency; native DPI scaling, custom hit-testing
AI — Analysis	Anthropic Claude Opus 4.8	Structured JSON output for backlog-level strategic analysis (default model, configurable via ANTHROPIC_MODEL_ANALYSIS)
AI — Scoring	Anthropic Claude Sonnet 4.6	Fast per-candidate RICE score suggestions (default model, configurable via ANTHROPIC_MODEL_SUGGESTIONS)
Auth & Data	Clerk + Supabase	Hosted auth with cloud-synced settings and feedback
Deploy	Vercel	Zero-config with serverless API routes for the Claude proxy

Architecture

flowchart LR
    subgraph Browser
      UI["React UI<br/>Next.js App Router"]
      LS[("localStorage<br/>guest mode")]
    end
    subgraph Vercel["Vercel — serverless"]
      AIAPI["AI API routes<br/>analyze · suggest-scores"]
      DataAPI["Data API routes<br/>workspaces · …"]
    end
    Clerk["Clerk<br/>auth"]
    Supa[("Supabase<br/>Postgres + RLS")]
    Claude["Anthropic Claude<br/>Opus · Sonnet"]

    UI -->|guest| LS
    UI -->|AI request| AIAPI
    UI -->|signed in| DataAPI
    AIAPI -->|hardened JSON calls| Claude
    DataAPI -->|verify session| Clerk
    DataAPI -->|service-role key| Supa

Loading

In guest mode the app is fully usable against localStorage. The AI API routes (analyze / suggest-scores) don't verify Clerk or touch Supabase — they read the request body and proxy Claude with ANTHROPIC_API_KEY so the key never reaches the browser. The data API routes (workspaces and related) are the ones that verify the Clerk session and talk to Supabase with the service-role key (RLS enabled as defense-in-depth).

Highlights

• Centralized scoring via useScored hook — RICE calculated once per state change, consumed by all components
• Memoized canvas positions — hover/selection interactions don't trigger position recalculation
• Responsive breakpoint via window.matchMedia hook — not CSS-in-JS or broken inline media queries
• Dual-mode AI — live Claude analysis via serverless proxy when available; smart demo fallback when not
• Serverless proxies — API keys stay server-side in app/api/analyze/route.js and app/api/suggest-scores/route.js

Getting Started

git clone https://github.com/kristenmartino/Tarazu.git
cd Tarazu
npm install
npm run dev

Enable Live AI Analysis (Local Development)

1. Copy .env.example to .env.local
2. Add your Anthropic API key (and any optional Clerk / Supabase / GA values)
3. Start the dev server:

   npm run dev

Next.js serves both the app and the /api/* routes from a single process at http://localhost:3000.

Production Deployment

Deploy to Vercel for automatic serverless function support (auto-detects the /api directory).

Without the API key, the app runs in demo mode with locally-generated analysis.

Deploy to Vercel

Add ANTHROPIC_API_KEY as an environment variable in Vercel's dashboard.

---

Project Context

Tarazu was built as part of a portfolio demonstrating PM + AI capabilities. The full PRD — including competitive analysis, requirements with acceptance criteria, technical architecture, risk mitigations, and launch plan — is available in docs/Tarazu_PRD.pdf. The complete brand system and product redesign spec lives in tarazu-brand-system-spec.md. The forward plan for the Validate feature is in docs/tarazu-validate-roadmap.md.

Built by Kristen Martino · Product & AI Strategist · MS Business Analytics & AI, UT Dallas