Domo AI Sales Engineer

Inspiration

The frustration of watching talented sales engineers spend hours delivering the same product demos over and over, while hot leads slip away outside business hours. We envisioned a world where every prospect could experience a perfect, personalized product demo instantly - whether it's 2 PM or 2 AM. DOMO was born from the belief that AI shouldn't just automate tasks; it should amplify human expertise and make it available 24/7.

What it does

DOMO allows users to create AI-powered product demos by uploading a knowledge base and a series of videos. A shareable URL is compiled that hosts a Tavus agent powered by this knowledge base that can intelligently answer prospective clients' questions as well as be able to intelligently discern when the user wants to see a demo video and display that video.

How we built it

We built DOMO using a modern full-stack architecture. The frontend is built with Next.js 13 (App Router), React, TypeScript, and Tailwind CSS. The backend is powered by Supabase, leveraging its PostgreSQL database, storage, and Edge Functions. We integrated Tavus CVI for a real time video ai agent and the backend uses the elevenlabs api to clone the voice and generate the text.

Challenges we ran into

We had a number of technical challenges we ran into. Our AI was defaulting to "product demo" responses when voice input couldn't be parsed properly. We solved this by implementing intelligent feature detection that recognizes keywords like "repository," "github," "database" and only triggers appropriate tool calls. Also, the AI's initial configuration was too restrictive with "ONLY tool calls" rules, causing awkward silences. We redesigned the persona to be conversational while maintaining function-focused behavior. Additionally, basic keyword matching failed for real-world queries with misspellings and synonyms. We built a sophisticated 4-layer search system that tries exact matching first, then falls back through increasingly fuzzy strategies. And finally, we had issues with multi-modal coordination. Synchronizing voice, video, and text responses across different APIs required careful state management and error handling to ensure a seamless user experience.

At this point the app is not functional due to difficulties with concurrent conversation issues with Tavus, but we are confident we can make it work.

Accomplishments that we're proud of

We are proud that with limited coding knowledge we were able to work as a team and resolve some of the technical issues that we had. One of our biggest accomplishments was being able to get our agent to make appropriate tool calls by engineering a resilient hybrid workaround on the client level. Instead of being blocked by external API limitations, we persisted and built a robust abstraction layer that makes our application resilient to unpredictable AI behavior. This breakthrough unlocked the core functionality of DOMO, enabling seamless video demos and knowledge base queries that make the AI feel truly intelligent and responsive rather than conversational but passive.

What we learned

We learned a lot when it comes to working with AI as well as working as a software development team.

AI Personas Need Personality: Pure function-focused AI feels robotic. The sweet spot is conversational intelligence that can still execute tools effectively.

Search is Harder Than It Looks: Real users don't type perfect queries. Building robust search that handles misspellings, synonyms, and partial matches is crucial for user experience.

Integration Testing is Everything: With multiple APIs (Tavus, Supabase, Netlify), comprehensive integration tests were the difference between a demo and a product.

Voice UI Changes Everything: When users can speak naturally, they expect natural responses. This pushed us to make our AI more intelligent and contextual.

Tool Calling is the Future: The ability for AI to take actions (play videos, query databases, trigger CTAs) transforms it from a chatbot into a true digital agent.

We also gained valuable experience in prioritizing features and adapting to changing circumstances under a tight deadline as our application did not end up being functional.

What's next for DOMO AI Sales Engineer

Semantic Search Integration: We'd like to leverage pgvector embeddings for even smarter knowledge retrieval Custom Avatar Training: This would allow companies to train DOMO on their specific product knowledge and brand voice

Advanced Analytics Dashboard: Something that could track prospect engagement, identify hottest leads, and optimize demo performance

CRM Integration: Being able to seamlessly connect with Salesforce, HubSpot, and other sales tools would expand the reach of our product and put it in the hands of as many prospective users as possible.

We believe that DOMO represents the future of sales enablement - where AI doesn't replace human expertise, but amplifies it to be available everywhere, every time. We're not just building a demo tool; we're creating the next generation of digital sales engineering.

Built With

• elevenlabs
• next.js
• react
• supabase
• tavus
• typescript