AI TRAFFIC ANALYST

Inspiration

I was inspired by the traffic 🚥 in Nigeria during rush hours in major cities like Enugu, Rivers, Lagos

What it does

AI Traffic Analyst is an intelligent system that: *Monitors real-time traffic conditions Analyzes historical traffic trends using SQLite Generates: 📊 Summaries of traffic patterns 📈 Traffic predictions *

How we built it

Backend:_ Python_ Database: SQLite3 (traffic + insights storage) LLM_Runtime:Llama.cpp LLM Engine: llama.cpp running lightweight models (TinyLlama ) Frontend: Streamlit UI Architecture: Traffic Data → SQLite → LLM Layer (Prompt Engine) → llama.cpp → Insights → UI Prompts are centrally managed in the LLM layer Streamlit handles only interaction + display LLM generates structured outputs using task-specific prompts

Challenges we ran into

I ran into many challenges like choosing the right gguf file for the project implementation of the file Type of prompt to send To Model Synchronizing real-time data ingestion with SQLite3 Connecting Streamlit UI with llama.cpp server reliably Displaying the LLM answers in Stremlit ** **Managing database paths and avoiding duplicate DB creation

Accomplishments that we're proud of

Built a fully local AI pipeline (no cloud dependency) 🔐 Designed a modular LLM architecture (separate UI, logic, DB) Implemented multi-mode intelligence: Summary Predictions Achieved real-time inference using lightweight models Created a system that mimics a traffic control room AI assistant

What we learned

Prompt engineering is critical for small LLM performance Separation of concerns (UI vs LLM vs DB) improves scalability Local LLMs require careful optimization and constraints Real_time data ingestion with SQLITE3