A multi-agent system for entrepreneurs to refine startup pitch presentations powered by LangGraph, LangChain, and Gemini Pro.
RECAP (Review Engine for Critiquing and Advising Pitches) is an LLM-powered agentic system designed to help founders and entrepreneurs receive actionable, multi-perspective, and structured feedback on their startup pitch presentations.
Leveraging Gemini Pro via langchain-google-genai, RECAP orchestrates multiple autonomous agents through LangGraph to simulate feedback from expert investors, mentors, and communication coaches.
Delivering a compelling pitch is vital for fundraising, networking, and market traction. However, high-quality, domain-relevant feedback is often:
- ❌ Expensive
- ❌ Inconsistent
- ❌ Inaccessible at scale
RECAP bridges this gap by delivering fast, repeatable, and context-aware AI evaluations — personalized for your niche and audience.
A multi-agent LLM system orchestrated through LangGraph for synchronous evaluation.
- 🧩 Autonomous agents for speech, content, and judgment.
- 🧠 Advanced prompt templates with expert prompting techniques.
- 🔄 Graph-based execution flow using
LangGraph. - 🧾 Typed feedback with
Pydantic v2schemas. - 🤖 Powered by Gemini Pro via LangChain's
langchain-google-genaiwrapper.
| Agent | Role |
|---|---|
🗣 SpeechAgent |
Evaluates clarity, confidence, pacing, and delivery. |
📄 ContentAgent |
Assesses business clarity, market opportunity, and scalability. |
🧑⚖️ JudgeAgent |
Emulates investor/judge perspective based on the target audience. |
🧷 Combiner |
Aggregates results and compiles cohesive final feedback. |
Each agent acts independently and their outputs are merged for a comprehensive review.
| Layer | Framework / Tooling |
|---|---|
| Language Model | Gemini Pro via langchain-google-genai |
| Orchestration | LangGraph for agent workflow management |
| Prompt Engineering | LangChain with structured PromptTemplate |
| Backend API | FastAPI for REST endpoints |
| Data Modeling | Pydantic v2 for schema enforcement |
| Environment Mgmt | uv |
| Deployment | Docker, Docker Compose, Uvicorn |
RECAP/
├── agents/ # LangChain agents (speech, content, judge)
├── core/ # State schema, config, and feedback combiner
├── api/ # FastAPI server entrypoint
├── pitch.py # CLI entry for local testing
├── pyproject.toml # Project metadata and dependencies
├── Dockerfile # Image specification
├── docker-compose.yml # Docker orchestration
└── README.md # Project documentation- Python 3.11+
uv- Google Gemini API Key
git clone https://github.com/SYED-M-HUSSAIN/RECAP.git
cd RECAP
# Create virtual environment
uv venv
source .venv/bin/activate
# Install all dependencies
uv pip install -e .
# Add your Gemini key
echo "GOOGLE_API_KEY=your-key" > .envStart the FastAPI server:
uvicorn api.main:app --reloadOpen the interactive API at: http://localhost:8000/docs
curl -X POST http://localhost:8000/evaluate \
-H "Content-Type: application/json" \
-d '{
"transcript": "Hi, I am Sarah from MedLink...",
"niche": "HealthTech",
"audience": "Investor"
}'Response includes structured feedback from all agents.
docker compose up --buildRuns RECAP inside a production-ready container with FastAPI and all agents activated.
- 🎙️ Speech-to-Text input via Whisper or Gemini Multimodal
- 🌍 Support for multilingual evaluation and regional investor expectations
- 📊 Performance dashboard with pitch history and iteration tracking
- 🔁 Integration with pitch practice platforms
- 🧠 Plugin-based custom agent extensions (storytelling, emotion, etc.)
This project is licensed under the MIT License.
See LICENSE for more details.
RECAP is built on top of world-class open-source libraries:
Author: Syed Muhammad Hussain Email: hs2764641@gmail.com