🚖 Uber Intelligence Platform

End-to-End Data Engineering & AI-Powered Analytics

An advanced data platform built with .NET Aspire, featuring a Medallion Architecture pipeline, Semantic Search with Vector Databases, and a Text-to-SQL Chatbot using local LLMs.

---

📸 Project Showcase (Visual Tour)

Note

Language Note: The UI is currently in Persian (RTL), but the data engine processes English datasets. Full English LTR support is easily configurable and planned for future updates.

---

🏗 System Architecture

The project leverages .NET Aspire to orchestrate a distributed environment without the complexity of manual configuration.

• Medallion Pipeline: Automated data flow from Bronze (Raw) → Silver (Cleaned) → Gold (Analytics) using Hangfire background jobs.
• Semantic Search: Vector embeddings generated via Ollama (all-minilm) and stored in Qdrant.
• AI Chatbot: Text-to-SQL engine powered by DeepSeek-Coder 6.7b to query the uber-db directly using natural language.

---

📂 Project Structure

• /Analyze: Exploratory Data Analysis (EDA). Includes analyze.ipynb and eda_charts.png.
• /IndexEffect: Performance benchmarking showing SQL execution plans before and after indexing.
• /Model: Ollama Modelfile for the custom uber-sql-helper LLM.
• /UberApp:
• frontend: Vite + React + Tailwind CSS dashboard.
• UberApi: .NET 10 Backend (Dapper, Hangfire, Qdrant Client).
• AppHost: Aspire Orchestrator (Api backend, Postgres and uber-db, Qdrant, Vite frontend).

---

🛠 Tech Stack

Component	Technology
Orchestrator	.NET Aspire
Backend	.NET 10, Dapper, Hangfire, Scalar UI
Frontend	React, Vite, Tailwind CSS
Databases	PostgreSQL (Relational), Qdrant (Vector)
AI Models	DeepSeek-Coder 6.7b (SQL), all-minilm (Embeddings)

---

🚀 Key Modules

📊 1. Analytics Dashboard

Real-time monitoring of 4 KPIs (Total Bookings, Success Rate, Revenue, Avg Rating) and 3 dynamic charts:

• Cancellation Distribution (Pie Chart).
• Vehicle Performance (Bar Chart: Counts with Ratings).
• Ride Traffic (Area Chart: Hourly trends with Min/Max/Avg stats).

⛓ 2. Medallion Pipeline & CSV Upload

Upload raw CSV datasets and watch the 3-stage transformation:

• Bronze: Raw data ingestion.
• Silver: Cleaning, trimming, and COALESCE handling.
• Gold: Final business-ready views.
• Managed by Hangfire with retry logic and history tracking.

🔍 3. Semantic Search (CRUD & Vector)

Manage ride records and search by "meaning" rather than keywords.

• Automatically syncs Postgres to Qdrant every 2 minutes.
• Uses Cosine Similarity to find related cancellation reasons.

💬 4. Chat with Database (Text-to-SQL)

Natural language interface for complex SQL queries.

• Local LLM generates SQL → Backend validates/executes → Frontend renders results.

---

⚙️ Setup & Installation

1. Prerequisites

• Ollama installed and running.
• .NET 10 SDK & Docker Desktop.

2. Prepare AI Models

ollama pull all-minilm
ollama pull deepseek-coder:6.7b
# Create the custom model using the Modelfile in /Model
ollama create uber-sql-helper -f ./Model/Modelfile

3. Run the App

Open UberApp.slnx and run the AppHost project. Aspire will automatically:

1. Spin up Postgres (with PgAdmin).
2. Spin up Qdrant.
3. Launch the Backend (Scalar UI at /scalar).
4. Launch the Frontend (Proxied via Vite to the API).