Demobot

Say goodbye to boring contact forms. Meet Charlie - your 24/7 AI sales assistant who talks to website visitors, qualifies leads, and books demos.

Comment

Inspiration

Contact forms and chatbots are where leads go to die.

After watching too many visitors bounce without converting, I realized we were overdue for a better way to qualify interest. When I saw Tavus’s Conversational Video Interface in the builder pack, it clicked: what if every website had a "Charlie", a 24/7 AI sales rep who never sleeps, never has a bad day and actually talks to prospects like a real human?

What it does

DemoBot replaces boring contact forms and chatbots with an AI sales assistant - "Charlie" - who engages website visitors in real-time. Charlie can:

  • Answer product questions 24/7
  • Qualify leads through conversation
  • Book demos with human sales reps via automatic Calendly integration
  • Surface performance insights via a powerful dashboard

The entire experience is built to feel natural, helpful, and seamless - no wait times, no form black holes. Visitors get instant, personalized responses. Teams get qualified leads delivered to their calendars. Everyone wins.

How I built it

I used the Bolt Hackathon Builder Pack tools to move fast:

Tavus CVI for the interactive AI video conversations with real facial expressions Daily's SDK to embed and manage real-time video communication JavaScript event handling to detect lead qualification and trigger automatic demo booking Netlify for rapid front-end hosting and deployment Bolt.new to scaffold and iterate on UI quickly (Planned: RevenueCat + Entri) for future monetization and customer onboarding

The key breakthrough was using Tavus tool calls to detect lead qualification events - enabling seamless, automated handoff to a human rep.

The demo site mimics a real SaaS company marketing page, with an embedded "Charlie" experience and a dashboard preview showing what customers would see if they used DemoBot.

Challenges I ran into

No-code builder, code-level problems: I’m not a developer, so stitching together APIs, SDKs, and frontend logic was a steep learning curve. Figuring out how the pieces fit without being able to read or write complex code made even small tasks take much longer.

Tavus conversation timeouts: Early on, Charlie would just... disappear. Conversations would end after 4 minutes, even if no one had interacted yet. It took time (and support messaged that frustratingly took days to get responses to) to understand how Tavus’s timeout system worked — and even more time to rebuild Charlie to create conversations dynamically, only when users engaged.

Cross-platform event integration: Getting iframe communication working between Tavus CVI and the parent page required careful JavaScript event handling that took days to figure out.

Bolt.new unpredictability: Bolt was powerful but unpredictable. Despite writing precise prompts, it would often rewrite entire sections of the app, remove working features (like Tavus), or restructure code unexpectedly. I spent entire days backtracking, undoing its changes, or fixing problems it created and at times had to compromise on polish or features just to avoid breaking the fragile core functionality.

Calendly integration headaches: One of the most frustrating (and ultimately satisfying) challenges was getting Calendly to appear at the right moment - only when Charlie had qualified a lead. This meant learning to listen for the correct tool call event from the Tavus iframe and triggering a popup at exactly the right time. It sounds simple now, but it took days of trial, error and testing.

Designing a useful dashboard: Most dashboards just show vanity metrics. I wanted ours to feel useful for someone running a sales team, surfacing actual insights about conversation performance, lead quality and business impact. Balancing realism with what was feasible during the hackathon was a constant tradeoff. had to compromise on polish or features just to avoid breaking the fragile core functionality.

Accomplishments that I'm proud of

End-to-end automation: Built a complete pipeline from conversation → qualification → booking that actually works live

Natural AI interaction: Charlie doesn’t just talk - he listens, qualifies and responds like a real SDR.

Multi-tool integration: Successfully combined multiple builder pack tools into one cohesive product

Real business impact: Created something that could genuinely replace contact forms and chatbots with measurable conversion improvements

Judge-ready demo: Anyone can test the live experience and see Charlie qualify them for a real meeting

What I learned

Small details matter: e.g., default timeouts in an AI video system can break your entire UX if you don't plan around them

AI agents need real product scaffolding around them: - good messaging, positioning, handoff flows and fallback logic

Fast tools are amazing but combining them thoughtfully with proper event handling matters more than just stacking them

Simple > complex. A working demo that shows value beats something half-baked but ambitious

What's next for Demobot

Build true customer onboarding with Entri and monetization via RevenueCat

Dynamic conversation creation: Start a fresh conversation for every visitor interaction to guarantee uptime

Multi-conversation ability: Allow as many simultaneous conversations as needed for true 24/7 uptime

Analytics dashboard: Surface real-time insights on lead quality, conversions and performance

CRM integrations: Pipe qualified leads directly into Salesforce, HubSpot and Pipedrive

Custom training: Let companies upload their own scripts and branding to train their version of Charlie

Multi-language support: Serve global websites with localized AI reps

Enterprise features: Team routing, lead scoring workflows and compliance controls

Launch a private beta with early adopters in the SaaS and agency space

Built With

  • ai-sales
  • bolt
  • calendly
  • conversation-design
  • cvi
  • daily-sdk
  • event-handling
  • iframe
  • javascript
  • lead-generation
  • low-code
  • netlify
  • no-code
  • react
  • tavus
  • video-ai
  • vite
  • webrtc
Share this project:

Updates