AI Application DevelopmentVoice AIGenerative AIMulti-Agent Systems

Pitch Fabrice

AI-Powered Startup Due-Diligence Interviewer

ProjectOverview

For a leading venture capital firm, early-stage founder screening requires significant manual effort. Pitch Fabrice automates this process through a voice-first AI platform that interviews founders, analyzes pitch decks, captures structured information and generates investor-ready debriefs.

The platform combines real-time voice interaction with specialized reasoning agents. A central Ledger maintains the startup data collected throughout the interview, while autonomous agents decide what to ask next, analyze responses and generate the final report.

TheChallenge

Early-stage screening involves repetitive interviews, pitch-deck analysis and manual documentation. The platform needed to automate this workflow while maintaining consistent, structured diligence.

Consistent, stage-aware interviews

Each founder requires different questions based on their startup stage and information already collected.

Reliable pitch-deck extraction

Uploaded decks must be classified and analyzed to extract structured startup information accurately.

Automated report generation

Interview responses need to become a consistent, structured investor debrief without manual preparation.

Reviewer feedback continuity

Human feedback must improve the generated debrief without restarting the review process.

Pitch Fabrice voice interview and debrief on desktop and tablet

The Solution

A multi-agent voice platform that interviews founders by stage and generates the debrief itself.

We designed Pitch Fabrice as a voice-first multi-agent platform, combining OpenAI Realtime for live conversations with AWS Bedrock agents for reasoning, extraction and analysis.

Founder responses and pitch-deck information flow into a central MongoDB Ledger. Planner and Analyzer agents continuously determine what information is missing, ask the next relevant question and update the Ledger.

After the interview, Claude Sonnet generates the structured investor debrief. Midas then sends it for human review, interprets reviewer feedback and regenerates the report until approval or the maximum review iterations are reached.

KeyFeatures

A voice-first diligence workflow, from the first question to the investor-ready report.

Conversational voice
interviews

Natural real-time founder conversations with low-latency responses and intelligent agent routing.

Multi-agent
orchestration

Specialized AI agents coordinate planning, questioning, analysis, confirmation and report generation.

Pitch deck
analysis

Vision AI extracts and cross-references information from uploaded pitch decks against founder responses.

Automated debrief
generation

Investor-ready reports are generated automatically from structured information collected during each interview.

Reviewer feedback
loop

Human reviewer feedback is interpreted and applied automatically to regenerate and refine the debrief.

TechStack

A real-time voice and agent stack, deployed on AWS with containerized delivery.

Next.js (TypeScript)

OpenAI Realtime API

OpenAI Agents SDK

AWS Bedrock

MongoDB Atlas

AWS App Runner

AWS S3

AWS Cognito

Microsoft Graph

Business Impact

More deal flow, consistent data.

Pitch Fabrice automates founder screening from pitch-deck analysis through interview and debrief generation. The investment team receives structured, consistently reviewed information from every conversation, reducing repetitive manual work while keeping human review in the final quality-control loop.

Get started