Skip to main content
Scribe Backend

​
Welcome to Scribe Backend

Scribe is a production-ready cold email platform built with FastAPI that uses a sophisticated multi-step AI pipeline to generate personalized academic outreach emails. The system combines web scraping, academic paper research, and Claude AI to create contextually relevant emails based on customizable templates.

​
Key Features

Multi-Step AI Pipeline

Four-stage agentic pipeline: template parsing, web scraping, academic enrichment, and email composition

Database-Backed Queue

Submit up to 100 email recipients at once with sequential processing and real-time status updates

Smart Web Scraping

Playwright-powered headless browser with JavaScript support for dynamic content extraction

ArXiv Integration

Automatic academic paper discovery and citation for research-focused outreach

Type-Safe Architecture

Pydantic models throughout with structured LLM outputs via pydantic-ai

Comprehensive Observability

Logfire integration with LLM call tracking for cost, tokens, and latency

​
Tech Stack

Scribe Backend is built with modern Python technologies:
  • Backend: FastAPI 0.109+, Python 3.13, Uvicorn
  • Database: PostgreSQL (Supabase), SQLAlchemy 2.0, Alembic
  • Authentication: Supabase Auth with JWT validation
  • AI/ML: Anthropic Claude (Haiku 4.5, Sonnet 4.5), pydantic-ai
  • Task Queue: Celery 5.3+, Redis 5.0+
  • Web Scraping: Playwright 1.56+, BeautifulSoup4, httpx
  • Observability: Logfire 4.14+ with auto-instrumentation

​
Quick Navigation

Quickstart

Get Scribe running in 5 minutes with step-by-step setup

Architecture

Understand the system design and deployment architecture

API Reference

Explore the complete REST API documentation

Pipeline Deep Dive

Learn about the 4-step email generation pipeline

​
Architecture Overview

┌─────────────────────┐
│  Template Parser    │  Claude Haiku extracts search terms & classifies type
│  (~1.2s)            │  Output: search_terms, template_type
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│   Web Scraper       │  Google Search + Playwright scraping + summarization
│   (~5.3s)           │  Output: scraped_content, urls, metadata
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  ArXiv Helper       │  Fetch academic papers (if RESEARCH type)
│  (~0.8s)            │  Output: arxiv_papers[]
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  Email Composer     │  Claude Sonnet generates final email
│  (~3.1s)            │  Output: final_email, writes to database
└─────────────────────┘

Total: ~10.4s

​
Template Types

Scribe supports three template types for different outreach scenarios:
  • Research: Includes research papers the professor has written
  • Book: Includes books the professor has published
  • General: General professional information about the professor
The pipeline automatically classifies your template and adjusts the enrichment strategy accordingly.

​
Production Deployment

The production backend is self-hosted on a Raspberry Pi with traffic routed through a Cloudflare Tunnel at https://scribeapi.manitmishra.com.

Health Check

GET /health - Check database connectivity and service status

API Documentation

GET /docs - Interactive Swagger UI for API exploration

​
Next Steps

1

Get Started

Follow the Quickstart guide to set up your local development environment
2

Understand Core Concepts

Learn about authentication, queue system, and template types
3

Explore the API

Check out the API Reference for detailed endpoint documentation
4

Dive into the Pipeline

Understand the 4-step pipeline that powers email generation

Build docs developers (and LLMs) love

Get started for freeTalk to us