Team Kautilyan

Instantly assemble a personal board of directors—AI mentors who answer in seconds, speak your language, and guide every big decision across chat, voice & video

Comment

Product Name: AI Advisory Board

Elevator Pitch (The one-liner for the project page):

AI Advisory Board is your on-demand council of experts. It's a multi-modal AI platform where you can tackle your toughest decisions by engaging with specialized personas through group chat, live voice calls, and 1:1 video sessions.

About the project

Inspiration

We live in an age of information overload, yet true wisdom feels scarcer than ever. Critical decisions—pivoting a startup, changing careers, or navigating a complex leadership challenge—are often made in isolation. Access to a "board of directors" or a diverse set of mentors is a privilege reserved for a select few.

Our inspiration was simple but profound: What if we could democratize high-level advisory guidance? What if anyone, anywhere, could instantly assemble a council of world-class experts—a strategist, a mentor, an innovator, and an empath—and have them debate their specific problem, right in front of them? We weren't inspired to build another chatbot; we were inspired to build an entire AI Advisory Board in the cloud.

What it does

AI Advisory Board is a multi-modal conversational AI platform that allows users to move seamlessly from insight to connection. It's not just about getting an answer; it's about understanding a problem from multiple, expert perspectives.

Here’s the journey the AI Advisory Board provides:

  • Assemble Your Council: Users start by visually selecting a team of AI advisors for a specific problem, creating either a focused 1:1 session or a dynamic group chat.
  • Witness the Debate: Our AI Orchestrator, powered by Gemini 1.5 Flash, facilitates a real-time text discussion between the selected advisors. They don't just answer in sequence; they ask clarifying questions, challenge each other's assumptions, and build on each other's points.
  • Hear Their Conviction: With a single click, users can escalate the text chat into a live group voice call. Powered by ElevenLabs' Conversational AI, each persona speaks in its own unique, pre-assigned voice, creating the feeling of a real, multi-participant roundtable discussion.
  • Go Deep with Video: For the most crucial conversations, users can launch into a 1:1 face-to-face video session with any advisor, powered by Tavus's real-time video replicas.
  • Build Your Own Board: Users aren't limited to our defaults. They can create and refine their own custom advisors, which are automatically assigned unique voices and personas, ensuring a truly personalized experience.

How we built it

We adopted a professional, UI-first development methodology, all orchestrated on Bolt.new.

  1. The Foundation (UI/UX): We started by designing and building the entire user interface in React, TypeScript, and Tailwind CSS. Our goal was a beautiful, intuitive, and production-worthy experience from day one. We used Framer Motion for fluid animations to make the application feel premium and responsive.
  2. The Backend (Supabase): The entire backend is powered by Supabase. We used Supabase Auth for secure, passwordless authentication. Our PostgreSQL database, designed with a robust schema and full Row-Level Security, serves as the single source of truth for users, advisors, and conversation history.
  3. The Brain (AI Orchestration): The core of AI Advisory Board is a Supabase Edge Function running our "Intelligent Orchestrator." This function uses the Gemini 1.5 Flash API with JSON mode to parse user queries, handle intelligent turn-taking, and facilitate clarifying questions and collaborative "chime-ins" between AI personas.
  4. The Voice (ElevenLabs): For the live voice calls, we integrated the @elevenlabs/client SDK. Our system uses a pool of 8 unique, pre-configured conversational agents. A master system prompt instructs each agent to act as a "voice actor," adopting the personality and instructions passed to it via dynamic variables for each conversational turn. This allows for seamless, multi-persona voice chat.
  5. The Face (Tavus): For 1:1 video, we integrated the Tavus CVI SDK. We maintain a pool of high-quality video replicas. When a user creates a custom advisor, our backend uses the Tavus API to programmatically create a new Persona (the brain) with the user's custom instructions and links it to one of our pre-made Replicas (the face).

Challenges we ran into

Our biggest challenge was a series of complex, real-world dependency and environment issues with the ElevenLabs integration. We were hit with ETARGET errors, versioning conflicts due to deprecated packages (elevenlabs vs. @elevenlabs/client), and finally, platform-level blockers where our Supabase Edge Function was unable to make outbound calls, likely due to a secret injection or firewall issue in the development environment.

This was our "darkest hour." However, we treated it like a real engineering team:

  1. We used Postman to systematically isolate the problem, proving our API key was valid.
  2. We dug through NPM documentation to find the definitive, correct package name and version.
  3. We made the pragmatic, strategic decision to implement a temporary client-side key workaround to ensure a fully functional demo, while documenting the production-ready, secure server-side architecture. This resilience and problem-solving under pressure is what we're most proud of.

Accomplishments that we're proud of

  • The Multi-Modal Flow: We are incredibly proud of the seamless journey from text to group voice to 1:1 video. This isn't just three separate features; it's one integrated experience that mirrors how human collaboration naturally escalates.
  • The Intelligent Orchestrator: Our Gemini-powered "Chief of Staff" AI is more than a simple router. The logic for asking clarifying questions and allowing advisors to intelligently "chime in" makes the conversations feel authentic and alive.
  • Dynamic Persona System: We successfully built a system where users can create their own custom advisors, which are then automatically and instantly brought to life with unique voices (from ElevenLabs) and personalities (for Tavus), creating a truly personalized experience.
  • Overcoming the Blockers: Debugging the dependency and platform issues felt like a real-world engineering challenge. Solving it under a tight deadline was a massive accomplishment.

What we learned

  1. APIs Evolve, Documentation is Truth: The world of AI SDKs is moving at lightning speed. We learned that you cannot trust outdated tutorials or assumptions. The official NPM page and API documentation are the only source of truth.
  2. Pragmatism Wins Hackathons: Our initial plan to dynamically create every asset via API was too complex and risky. We learned to make the hard decision to switch to a pre-configured asset pool, which massively de-risked the project and improved reliability without sacrificing the core user experience.
  3. Isolate, Don't Assume: When faced with a bug, our most effective tool was isolation. Using Postman to test the API key directly saved us hours of debugging the wrong part of the stack.

What's next for Team Kautilyan

The AI Advisory Board is more than a hackathon project; it's the beginning of a new paradigm in personalized guidance. Our roadmap includes:

  • Conversation Memory & Relationship Building: Implementing a vector database (like Supabase pgvector) to give advisors long-term memory, allowing them to remember past conversations and build a real, evolving relationship with the user.
  • Enterprise Tier: A version for teams where companies can create a board of directors for their employees, trained on internal company knowledge.
  • Proactive Nudges: Allowing advisors to proactively reach out to the user based on their stated goals (e.g., "Hi Sriharsha, you mentioned you wanted to work on your leadership skills. Here's a quick thought for your week...").

Built with

  • Languages: TypeScript, SQL
  • Frameworks: React, Vite, Node.js
  • Platforms: Bolt.new, Supabase
  • Frontend: Tailwind CSS, Framer Motion, Lucide React
  • Cloud Services: Supabase Auth, Supabase Functions, Supabase Postgres
  • APIs:
    • Google Gemini 2.0 Flash API (for AI orchestration)
    • ElevenLabs Conversational AI API (for dynamic, multi-persona voice calls)
    • Tavus CVI & Replicas API (for 1:1 real-time video sessions)

Built With

Share this project:

Updates