AI-powered sales intelligence that captures, analyzes, and syncs call data to your CRM automatically.
Built by Prema Vision LLC, an AI automation consultancy led by Denys Korolkov.
🎬 Live Demo (2 min)
This short walkthrough shows how a sales call is uploaded, transcribed, analyzed, and converted into structured CRM-ready insights — with a human always in control.
- Web UI: https://prema-sales-call-summarizer.streamlit.app
- API Docs (Swagger): https://prema-sales-call-summarizer.onrender.com/docs (Note: API docs availability depends on deployment configuration)
Sales teams lose 60% of critical customer context the moment a call ends. Prema Sales Call Summarizer is an automated intelligence layer that ingests audio, extracts high-value business insights, and syncs structured data directly to your CRM. It eliminates manual data entry and ensures risk factors and action items are captured accurately. This solution bridges the gap between conversation and system of record.
This project demonstrates how AI can operationalize sales intelligence without requiring changes to the team's existing workflow. For high-velocity sales and customer success teams, data integrity is the bottleneck to scaling.
- Sales Ops: Eliminates "shadow CRM" where data lives in reps' heads.
- RevOps: Provides structured data for objection handling and product feedback loops.
- Management: Offers auditable, AI-generated summaries for coaching and deal reviews.
- Multi-Format Audio Ingestion: Supports MP3, WAV, M4A via secure upload.
- Enterprise-Grade Transcription: Utilizes OpenAI Whisper for near-human accuracy.
- Context-Aware AI Analysis: Extracts executive summaries, key pain points, and objection handling patterns.
- Automated Follow-up: Generates draft emails based on call context.
- CRM Sync Layer: Pushes structured data to CRM (stubbed for demo, architecture-ready for HubSpot/Salesforce).
- Dual Interface: Robust REST API for integrations + dedicated Streamlit dashboard for manual usage.
The system follows a modular service-oriented architecture designed for scalability and maintainability. The architecture separates ingestion, transcription, analysis, and CRM sync into modular services, enabling easy scaling and replacement of components.
graph TD
subgraph Frontend
UI[Streamlit Dashboard]
API_Client[External API Client]
end
subgraph Backend ["FastAPI Backend"]
Router[API Routes]
subgraph Services ["Core Services"]
CallSvc[Call Service]
TransSvc[Transcription Service]
AnalSvc[Analysis Service]
CRMSvc[CRM Sync Service]
end
Storage[Audio File Storage]
DB[(PostgreSQL / SQLite)]
end
subgraph External ["External Providers"]
OpenAI["OpenAI API<br/>(Whisper & GPT-4)"]
CRM["CRM System<br/>(HubSpot / Salesforce)"]
end
UI -->|HTTP Requests| Router
API_Client -->|HTTP Requests| Router
Router --> CallSvc
Router --> TransSvc
Router --> AnalSvc
Router --> CRMSvc
CallSvc -->|Read/Write| DB
CallSvc -->|Save| Storage
TransSvc -->|Audio| OpenAI
TransSvc -->|Update| DB
AnalSvc -->|Transcript| OpenAI
AnalSvc -->|Update| DB
CRMSvc -->|Push Data| CRM
CRMSvc -->|Log Status| DB
graph TD
A[User/Client] -->|Upload Audio| B(FastAPI / Streamlit)
B -->|Save File| C[Audio Storage]
B -->|Create Record| D[(Database)]
subgraph AI Pipeline
E[Transcription Service] -->|Send audio| F[Whisper Model]
F -->|Transcript| E
G[Analysis Service] -->|Transcript| H["LLM (GPT-4)"]
H -->|Insights| G
end
B -->|Trigger| E
E -->|Update DB| D
B -->|Trigger| G
G -->|Update DB| D
subgraph Integrations
I[CRM Service] -->|Sync Data| J[External CRM]
end
B -->|Sync| I
I -->|Log Status| D
- Backend: Python 3.11, FastAPI, Pydantic
- Data Layer: SQLModel (SQLAlchemy), SQLite (Dev), PostgreSQL (Prod)
- Frontend: Streamlit
- AI Services: OpenAI Whisper (ASR), GPT-4o-mini (LLM)
- Testing: Pytest, Playwright (E2E)
- Deployment: Render (Docker/Python runtime)
- Python 3.11+
- OpenAI API Key (optional for stub mode)
# Clone the repository
git clone <repo-url>
cd prema-sales-call-summarizer
# Create virtual environment
python3.11 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
# Install dependencies
pip install --upgrade pip setuptools wheel
pip install -r requirements.txt
# Environment Setup
cp .env.example .env
# Edit .env to add your OPENAI_API_KEYOption 1: Dedicated Dashboard (UI)
streamlit run app/ui/streamlit/dashboard.pyAccess at http://localhost:8501
Option 2: REST API (Backend)
uvicorn app.main:app --reloadDocs at http://localhost:8000/docs
- Upload: Navigate to the "Upload New Call" sidebar. Enter call metadata and upload an audio file.
- Transcribe: Click the 🎙️ Transcribe button on the call card. The system processes the audio and generates a verbatim transcript.
- Analyze: Click 🧠 Analyze. The LLM extracts insights, pain points, and drafts a follow-up email.
- Review & Edit: Edit the generated email draft if needed. Select specific "Action Items" to track.
- Sync: Click 🔄 Sync CRM. Data is pushed to the CRM system (simulated in this demo) and the status updates to
SYNCED.
Create a Call
curl -X POST "http://localhost:8000/api/v1/calls/" \
-F "title=Discovery with Acme" \
-F "[email protected]"Trigger Analysis
curl -X POST "http://localhost:8000/api/v1/calls/{call_id}/analyze"- CTOs & Engineering Leads: Looking for a reference architecture for AI-driven workflow automation.
- Founders: Evaluating the feasibility of custom sales AI tools.
- Sales Operations: Seeking to automate CRM data entry and improve data quality.
- Real-time Processing: Streaming transcription for live call assistance.
- Vector embeddings (optional integration-ready): Querying past calls for competitive intelligence.
- Native Integrations: Direct OAuth connections to HubSpot, Salesforce, and Pipedrive.
- Multi-Tenant Support: Architecture supports schema isolation for SaaS deployment.
Denys Korolkov — Prema Vision LLC