Assistant

Project Details

Assistant is a personal Telegram bot built for experimentation and learning around agentic workflows, background processing, and finance-focused utilities. It runs as an ASP.NET Core webhook API and combines natural-language chat, reminder and task handling, live TEFAS fund analysis, and credit card statement PDF analysis in a single Telegram-first interface. Agent behaviors are implemented with Microsoft Agent Framework, with custom personality, memory, and task-awareness layers tailored for Telegram conversations.

Key Features

  • /chat runs a tool-enabled conversational agent that keeps per-user context, understands natural-language reminder requests, and can pull in fresh web information when the question depends on recent facts.
  • /expense shows the current expense summary and can analyze uploaded credit card statement PDFs.
  • /tefas <FUND_CODE> fetches live TEFAS data, normalizes fund metrics, and returns an AI-assisted summary with a deterministic fallback.
  • Incoming Telegram updates are validated, queued, and processed asynchronously through Hangfire jobs.

Technical Implementation

  • Built with ASP.NET Core 10, Telegram.Bot, PostgreSQL, and Hangfire using a webhook-driven architecture.
  • Designed an extensible command pipeline around IBotCommand, BotCommandFactory, and a centralized update handler.
  • Used Microsoft Agent Framework to structure agent execution, tool-enabled interactions, and conversational flows.
  • Configured the chat agent with personality injection, saved user memories, pending-task context, current-time resolution, web search, and one-time or recurring task scheduling tools.
  • Preserved conversation continuity with per-chat sessions and chat history summarization so the agent can carry forward open loops, decisions, and user preferences across longer Telegram threads.
  • Integrated Google Gemini for chat, reminder interpretation, expense analysis, and fund summarization.
  • Added document processing by sending uploaded PDFs to Markitdown, extracting statement text, and persisting billing-period expense summaries.
  • Implemented live TEFAS scraping and HTML parsing to convert fund pages into structured snapshots before AI summarization.
  • Secured the webhook flow with Telegram secret-token validation and optional allowed chat ID checks.

Project Link

View Next

University Announcement Tracker

A mobile and web application that allows users to follow announcements from their chosen universities. It enables users to list, share, and access details of announcements and sends notifications for new updates.