Mirror

Every founder validation method is structurally biased toward false positives. Interviews reward founders who can pitch. Surveys attract respondents who want to be helpful. Accelerator feedback comes from advisors optimizing for encouragement. By the time you discover real market resistance, you've spent six months building in the wrong direction.

Mirror is the feedback loop that doesn't have a social contract. 200+ AI personas with no incentive to be kind, no scheduling, no filter. Submit your idea in plain language — in under 60 seconds, Mirror tells you who wants it, who doesn't, and exactly why.

A live simulation dashboard showing reactions from personas matched to your product niche — each with their own demographics, income bracket, risk tolerance, and communication style. You see sentiment scores, specific objections, and geographic distribution mapped on a real-time globe. You can filter by segment, drill into individual persona reactions, and re-run with a refined idea.

You can also call a persona. Mirror includes a live voice agent — have a real conversation with a synthetic customer, push back on their objections, pitch them differently, and hear how they respond in real time.

200+ persona profiles embedded with Cohere embed-v3 — each encoding age, income bracket, risk tolerance, brand affinities, communication style, and psychographic markers as structured parameters stored in a vector index. A Cohere reranker scores every profile against the submitted idea and returns only the niche-relevant subset.

A consumer wellness app won't hear from enterprise procurement managers. A developer tool won't be evaluated by demographics who've never seen an IDE. The retrieval layer is what makes the simulation feel targeted rather than random.

The generation pipeline injects persona attributes as typed system context — not freeform roleplay instructions, but structured parameters the model uses to produce grounded, persona-specific objections. The core engineering challenge: AI is agreeable by default. Getting 200 personas to disagree with each other authentically — not just randomly — required iterative prompt architecture that went significantly beyond standard prompt engineering.

The simulation dashboard streams reactions via Supabase Realtime and updates each persona's sentiment dot on a Three.js globe live as reactions resolve. The model is the renderer; the persona profile is the scene.

Mirror includes a real-time voice agent layer powered by Vapi's WebRTC integration. You call a persona, have a live conversation, and hear exactly how they'd respond to your pitch. The harder engineering problem was stateful persona memory: the persona must hold its stated opinions under pressure, remember what it said earlier in the call, and stay in character throughout. Stateful coherence across a live call is a different system design problem from a text pipeline.

Built on Next.js 15 App Router with React 19 and TypeScript throughout. Persona vectors stored in a Cohere vector index; Supabase Realtime streams simulation results to the dashboard as each persona resolves. Three.js renders the live sentiment globe. Auth is a dual layer: Auth0 for user management, Supabase Auth for session handling. ReactFlow powers the idea canvas for structuring and iterating on ideas with evidence and scout nodes.

Mirror is live and accepting waitlist signups. If you're a founder, PM, or anyone who needs honest market feedback before committing to a build — Mirror is built for you. Planned features include industry-specific persona cohorts, longitudinal tracking across idea iterations, team collaboration on simulations, and shareable report formats for investor conversations.