← Projects

What I did

Deneb

A multi-agent AI backend behind a Telegram bot

Deneb is a full-stack personal finance tracker. A Telegram bot receives the input. The bot reads text, photos, PDFs, and voice notes. A root agent routes each request to a specialist agent. The agents save and search expenses in MongoDB. The visualizer agent generates charts with Matplotlib.

Google ADKPythonaiogramMongoDBTelegram BotMatplotlib
View on GitHub →

Features

  • The root agent routes each request to a specialist agent.
  • The saver agent stores expenses in MongoDB.
  • The retriever agent searches expenses with filters.
  • The visualizer agent generates charts with Matplotlib.
  • The bank agent reconciles expenses with bank statements.
  • A YOLO model crops receipt photos for text recognition.

How it works

  1. Send a message or a file to the Telegram bot.
  2. Let the root agent route the request.
  3. Read the expense that the saver agent recorded.
  4. Ask the bot to show or search your expenses.
  5. Ask the bot to generate a chart.

Tech stack

  • Backend: Python 3.13, Google ADK, aiogram
  • Database: MongoDB, Beanie, Motor
  • Models: Gemini 2.5 Flash, Gemini 3 Flash Preview
  • Visualization: Matplotlib, Pandas
  • OCR: YOLO, ONNX Runtime
  • Observability: AgentOps