AI Project Tracker (Self-Hosted LLM) 🧠

Task management system powered by a self-hosted LLM.

This project combines a Telegram bot, web dashboard, and FastAPI backend to generate project roadmaps and track progress.
Runs entirely offline using a local LLM (Qwen 2.5 via llama.cpp) — no external APIs required.

---

🚀 Overview

• Generates structured project roadmaps from user input using LLM
• Supports both Telegram bot and web interface
• Runs fully offline (self-hosted AI, no OpenAI dependency)
• Built with async backend and containerized architecture
• Designed for privacy, control, and low-latency inference

---

🏗 Architecture

graph TD
    User((User))
    TG[Telegram Bot]
    Web[Web Dashboard]
    API[FastAPI Backend]
    DB[(PostgreSQL)]
    Redis[(Redis Cache)]
    LLM[Local LLM Service]

    User --> TG
    User --> Web
    TG --> API
    Web --> API
    API --> DB
    API --> Redis
    API -->|Prompt| LLM
    LLM -->|Stream| API

Loading

---

🚀 Key Features

Self-Hosted AI

Runs quantized LLM models (GGUF) locally via llama.cpp No OpenAI API keys required

Task Generation

Transforms user ideas into 5–7 structured technical tasks

Multi-Platform Access

• Telegram Bot (Aiogram 3)
• Web Dashboard (Next.js)

Real-Time Feedback

Streaming responses and progress updates

---

🛠 Tech Stack

Backend

• FastAPI
• SQLAlchemy (async)
• Pydantic

Frontend

• Next.js 14
• TypeScript
• Tailwind CSS

AI

• llama.cpp (OpenAI-compatible API)
• Qwen 2.5 (GGUF models)

Infrastructure

• Docker / Docker Compose
• Redis
• PostgreSQL
• Caddy

---

⚡ Quick Start

Clone and configure

git clone https://github.com/vlimkv/ai-project-tracker.git
cp .env.example .env

Run system

# Downloads model (~2GB on first run)
docker compose up -d --build

---

🌐 Access

• Web UI: http://localhost:3000
• API Docs: http://localhost:8000/docs
• LLM Endpoint: http://localhost:8080/v1

---

⚙️ Environment

• AI_PROVIDER — oss (local) or openai
• OSS_MODEL — qwen2.5-3b-instruct
• REDIS_URL — caching and state

---

⚠️ Requirements

• Docker
• 4GB RAM minimum (8GB recommended)
• GPU optional

---

💡 Why this project

Built to explore real-world LLM integration without external APIs.

Focus:

• self-hosted AI
• async backend
• full-stack system design

---

📌 Notes

This project demonstrates:

• local LLM deployment
• AI + backend integration
• full-stack architecture (bot + web + backend)
• containerized development workflow