PitchGaunlet

Voice AI that validates startup ideas in 90 seconds. Get YC-level honest feedback before wasting 6 months building the wrong thing. Pivot faster, win sooner.

Comment

Inspiration

Every founder has that moment: you've spent 6 months building your startup idea, only to discover the market is saturated, customers don't want it, or someone already solved it better. We've seen too many brilliant founders waste time and money on ideas that were doomed from the start.

The problem? Getting honest, YC-level feedback requires connections, money, or luck. Most founders never get the brutal truth they need before committing months to building.

We built PitchGaunlet to democratize startup validation. What if every founder could get brutally honest, data-driven feedback in 90 seconds? What if they could learn from thousands of historical startup outcomes before writing a single line of code?

What it does

PitchGaunlet is a voice-powered AI validation platform that gives founders YC-level market feedback in 90 seconds. Here's how it works:

  1. Voice Conversation: Founders naturally describe their idea to an AI agent (powered by Vapi)
  2. Intelligent Analysis: GPT-4 extracts structured data (industry, problem, customers, keywords)
  3. Historical Intelligence: Redis vector search finds similar historical startup ideas and calculates success rates
  4. Deep Market Research: Parallel API scrapes the web for real competitors, customer complaints, and hiring signals
  5. Brutal Assessment: Claude Sonnet 4 provides honest feedback with 7 key sections:
    • Brutal Truth
    • Market Reality
    • Fatal Flaws
    • Hidden Opportunities
    • Concrete Next Steps
    • Go/No-Go Recommendation
    • Pivot Ideas
  6. Shareable Reports: Beautiful, comprehensive reports stored in Sanity CMS with shareable URLs

The result? Founders get actionable insights that would normally take weeks of research and thousands of dollars in consulting fees—all in under 90 seconds.

How we built it

Frontend:

  • Next.js 16 with React 19 and TailwindCSS for a modern, responsive UI
  • Real-time progress tracking with step-by-step analysis visualization
  • Dynamic report viewer with beautiful typography and layout

Voice AI:

  • Vapi for natural voice conversations
  • Automatic transcript generation

AI Pipeline:

  • OpenAI GPT-4 for structured data extraction from transcripts
  • OpenAI text-embedding-3-small (1536 dimensions) for semantic similarity
  • Anthropic Claude Sonnet 4 for comprehensive, honest assessments

Vector Database:

  • Redis with RedisVL for AI-powered similarity search
  • HNSW algorithm with cosine similarity
  • Stores historical startup ideas with embeddings for pattern matching

Market Research:

  • Parallel API for real-time web scraping
  • Competitor discovery from live web searches
  • Customer complaint analysis from Reddit, Twitter, G2
  • Hiring signal detection from job boards

Data Storage:

  • Sanity CMS for beautiful, shareable reports
  • Structured schema with all analysis data
  • Dynamic routing for report URLs

Architecture: The system uses a modular pipeline architecture where each step can run independently or as part of an orchestrated flow. We implemented graceful error handling so the system degrades elegantly if any service is unavailable.

Challenges we ran into

  1. Parallel API Integration: Initially used the wrong endpoint and authentication method. We spent hours debugging DNS errors before discovering the correct beta endpoint (/v1beta/search) and header format (x-api-key instead of Bearer token).

  2. Vector Search Implementation: Getting Redis vector search working required understanding HNSW algorithms, embedding dimensions, and binary vector storage. We had to convert JavaScript arrays to FLOAT32 buffers for Redis compatibility.

  3. Real-time Data Extraction: Parallel API returns unstructured web data. We built sophisticated extraction logic to clean excerpts, extract meaningful content, and handle edge cases like empty results or malformed data.

  4. Graceful Degradation: The system integrates 6+ external services. We implemented comprehensive fallback mechanisms so the platform works even if Redis, Parallel API, or other services fail—ensuring users always get value.

  5. Cost Optimization: With multiple AI API calls per validation, costs could spiral. We optimized by using cheaper embedding models, implementing caching strategies, and designing efficient prompts to minimize token usage.

  6. Competitor Description Quality: Initial fallback system generated generic templated descriptions. We built a curated database of 16+ major companies with real descriptions, plus intelligent industry-aware fallbacks.

Accomplishments that we're proud of

  1. Complete End-to-End Pipeline: Built a production-ready system that takes voice input and produces comprehensive validation reports—all in 90 seconds.

  2. Real-Time Web Research: Successfully integrated Parallel API to scrape live competitor data, customer complaints, and hiring signals from the web in real-time.

  3. AI Vector Search: Implemented sophisticated Redis vector similarity search that finds similar historical startup ideas and calculates success rates using 1536-dimensional embeddings.

  4. Brutal Honesty: Created an assessment system that provides genuinely useful, actionable feedback—not just generic encouragement. Claude's assessments identify fatal flaws, hidden opportunities, and concrete pivot ideas.

  5. Beautiful UX: Designed an intuitive interface with real-time progress tracking, so users understand exactly what's happening at each step of the analysis.

  6. Cost Efficiency: Achieved ~$0.28-0.85 per validation by optimizing API usage, using efficient models, and implementing smart caching.

  7. Production-Ready Architecture: Built with modularity, error handling, and scalability in mind. The system gracefully handles failures and can be extended with new data sources or AI models.

What we learned

  1. Voice AI is Powerful: Natural voice conversations capture nuance that text forms miss. Founders express ideas more naturally when talking, leading to better analysis.

  2. Vector Search is Game-Changing: Semantic similarity search on historical data provides insights that keyword matching never could. Finding "similar" ideas based on meaning, not just keywords, revealed patterns we didn't expect.

  3. Real Data > Mock Data: Initially using mock data for competitors felt fine, but switching to real Parallel API results made the assessments dramatically more valuable. Real customer complaints and actual competitor data changed the quality of feedback.

  4. Graceful Degradation is Essential: When building with 6+ external APIs, things will fail. Building robust fallbacks from day one saved us during demo time when services had hiccups.

  5. Cost Management Matters: AI APIs are expensive. We learned to optimize prompts, use cheaper models where appropriate, and implement caching—reducing costs by 60% without sacrificing quality.

  6. Brutal Honesty is Valuable: Users actually want harsh feedback. Being honest about market saturation, competition, and fatal flaws is more valuable than encouragement—and founders appreciate it.

  7. 90 Seconds is the Sweet Spot: Fast enough to feel instant, slow enough to build trust that real analysis is happening. This timing creates the perfect user experience.

What's next for PitchGaunlet

Short-term (Next Month):

  • PDF export for reports
  • Email notifications when analysis completes
  • Social sharing capabilities
  • User accounts to save and compare multiple ideas

Medium-term (Next Quarter):

  • Integration with real Twitter and Reddit APIs (beyond Parallel API scraping)
  • Crunchbase integration for funding data
  • Product Hunt integration for competitor discovery
  • Advanced analytics dashboard showing validation trends

Long-term Vision:

  • AI-Powered Pivot Suggestions: Not just identifying problems, but generating concrete pivot ideas with market validation
  • Founder Community: Connect founders with similar ideas to share learnings
  • Investor Network: Allow investors to browse validated ideas and connect with promising founders
  • Success Tracking: Follow up with founders to track which validations led to successful startups
  • Industry Reports: Aggregate data to publish insights on startup trends and success patterns

Technical Improvements:

  • Implement response caching to reduce API costs by 40%
  • Add more historical startup data (targeting 1000+ ideas in Redis)
  • Build a recommendation engine that suggests similar ideas to explore
  • Add multi-language support for global founders

The ultimate goal: Make PitchGaunlet the go-to platform for startup validation, helping thousands of founders avoid wasting time on doomed ideas and focus on building what actually matters.

Built With

  • claude
  • openai
  • parrallel
  • react
Share this project:

Updates