AI Telegram Scheduler Bot

A production-ready backend service for parsing news, generating AI-powered posts, and scheduling Telegram messages for delivery at a specified time.

Designed with reliability, atomic execution, and future scalability in mind.

---

📜 License

This project is licensed under a Proprietary Non-Commercial License.

You are allowed to view and evaluate the source code for personal or educational purposes only.

❌ Commercial use, redistribution, or reuse of the core logic without explicit written permission is strictly prohibited.

© 2026 Arslonbek Erkinov. All rights reserved.

---

🚀 Features

• 📰 News parsing from external sources
• 🤖 AI-powered post generation
• ✍️ Automatic content preparation for Telegram
• 📆 Schedule Telegram messages for a specific datetime (UTC-based)
• 🔒 Atomic database locking (no duplicate delivery)
• 🔁 Jobs survive application restarts
• 🧠 Clear job lifecycle: scheduled → sent / failed
• 🗄 SQLite database with WAL mode enabled
• 📜 Structured logging for observability
• ⚙️ Clean architecture (API / Services / Repositories / Scheduler)

---

🧠 Core Capabilities

📰 News Parsing

The system can fetch and parse news data from external sources (RSS / APIs / scrapers), normalize content, and prepare it for further processing.

Parsed data is stored and can be:

• reviewed
• enriched
• passed to AI for generation

---

🤖 AI Post Generation

An integrated AI layer is used to generate Telegram-ready posts:

• Headline rewriting
• Summary generation
• Tone adaptation (news / neutral / promotional)
• Emoji & formatting optimization
• Language-agnostic (model-based)

AI generation is decoupled from scheduling logic and can be extended independently.

---

📆 Smart Scheduling

Generated (or manual) posts can be scheduled for future delivery with:

• Timezone-safe UTC normalization
• Validation against past timestamps
• Automatic job recovery after restart

---

🏗 Architecture Overview

API (FastAPI)
│
├── News Parser
│
├── AI Content Generator
│
▼
Scheduler Service
│
▼
SQLite Database (WAL, Atomic Lock)
│
▼
APScheduler
│
▼
Telegram Bot API

🧠 Key Design Decisions

Atomic Execution

Each scheduled post is protected by an atomic database lock, ensuring:

• No duplicate sends
• Safe concurrent execution
• Idempotent job processing
• Protection against race conditions

This design mirrors row-level locking patterns used in PostgreSQL.

---

SQLite + WAL

SQLite was intentionally chosen for the first stage:

• Reliable for single-instance schedulers
• WAL mode allows concurrent reads/writes
• Easy migration to PostgreSQL in the future

APScheduler

• Precise job execution
• Jobs persist in memory and survive restarts
• Clean integration with Python services

---

📦 Tech Stack

• Python 3.11+
• FastAPI
• APScheduler
• SQLAlchemy
• SQLite (WAL mode)
• Aiogram (Telegram Bot API)
• AI provider (OpenAI / local LLM ready)

📂 Project Structure

app/
├── api/            # HTTP API endpoints
├── services/       # Business logic
├── scheduler/      # APScheduler jobs & tasks
├── repositories/   # DB access layer
├── models/         # SQLAlchemy models
├── database.py     # DB engine & session
├── config.py       # Settings (.env)
├── telegram/       # Telegram client
├── main.py         # FastAPI entrypoint
logs/

---

🔌 API Example

Schedule a message

POST /api/schedule

{
  "text": "Hello from the scheduler 🚀",
  "publish_at": "2026-01-06T18:00:00+03:00"
}

Response

{
  "status": "scheduled",
  "publish_at": "2026-01-06T15:00:00+00:00"
}

🧪 Reliability Guarantees

• A message is sent only once
• If execution fails:
• Status is marked as failed
• Error is stored in DB
• Duplicate execution attempts are automatically blocked

🛠 Future Improvements (Planned)

The following enhancements are intentionally prepared but not enabled yet:

• 🐘 PostgreSQL support (row-level locking, job store)
• 🐳 Docker & Docker Compose
• 🔴 Redis (distributed locks & queues)
• 📊 Admin dashboard
• 🔁 Retry backoff strategies
• ☁️ Horizontal scaling