JAZO - Interview Bot

Inspiration: Capturing authentic marketing content, such as founder stories, customer case studies, or podcast segments is incredibly difficult. People hate staring at blank pages, and talking to standard AI chatbots often feels like a sterile, robotic interrogation. We wanted to build an interviewer that is genuinely worth talking to. JAZO was born out of the idea that an embodied, charismatic AI, featuring an expressive face and a warm personality would break down walls and make people actually want to share their stories. We wanted to make the interview experience personal, fun, and unique so that users would gladly come back to JAZO again and again.

What it does: JAZO is a fully agentic hardware and software interviewer designed to capture rich marketing narratives. You brief JAZO exactly like an editor briefs a reporter: who is being interviewed, the topic, and the needed content.

JAZO then takes over, conducting a live, voice-driven conversational interview. But JAZO isn't just a voice, it features a dynamic, expressive LCD face that reacts to the conversation with emotions like empathy, joy, and curiosity.

Once the interview wraps up, JAZO acts as a local MCP (Model Context Protocol) server. Instead of just dumping a raw transcript, JAZO intelligently quiries the interview context, search the transcript, and extract verified, word-for-word pull quotes to draft a highly polished marketing asset.

How we built it The Character & UI: We built a custom Next.js frontend and an ESP32-controlled LCD face. The backend continuously analyzes the conversation to determine JAZO's mood, updating the physical/web face in real-time to match the emotional tone of the user's story. The Brains: We utilized Gemini with strict personality constraints to ensure JAZO remains warm, human, and fiercely curious, willing to abandon a rigid script when the user says something fascinating.

Challenges we ran into Arduino JSON Schema Desync: We tried adding a new "sad" emotion to Jazo's face, but this immediately crashed the physical Arduino hardware because the C++ firmware was strictly typed to a specific JSON schema of moods. Instead of flashing new firmware mid-demo, we had to rely on strict LLM prompt engineering, instructing the model to map the user's request for "sadness" to the hardware's existing "empathetic" state to prevent a total desync.

Accomplishments that we're proud of Embodiment: We didn't just build a software tool; we built a character. JAZO's face completely changes the dynamic of the interview, turning it from an interrogation into a friendly, engaging chat. Zero-Hallucination Quotes: Successfully implementing the MCP Server allows the AI to sift through the transcript and only extract verified, word-for-word pull quotes instead of hallucinating them from memory. High Agency & Speed: We maintained relentless momentum throughout the event, iterating on our architecture mid-build to ensure our product perfectly fit the Prologue Stories bounty requirements. We constantly seeked custom reviews and iterated based on the reviews we got.

What we learned The Power of Embodiment: We learned that people open up dramatically more when they have a visual, reactive face to talk to. Hardware and personality go a long way in AI. Prompt Engineering: We learned the extreme nuance required to keep an AI interviewer warm and curious, rather than rigidly adhering to a checklist of questions.

What's next for JAZO - Interview Bot Dedicated Hardware: Refining the physical hardware body for JAZO to sit on desks as a dedicated "Marketing Reporter" for startup teams and agencies. Prologue Integration: Integrating JAZO directly into Prologue's "Premise" software roadmap to help scale authentic story collection for businesses everywhere.