Traffic congestion is a critical challenge in rapidly urbanising cities, and conventional fixed-time traffic signals fail to adapt to dynamic real-time variations, leading to longer waiting times, fuel wastage, emissions, and delays in emergency response. To address this, the project designs and implements an AI-Driven Adaptive Traffic Signal Control System at a six-road intersection near Adyar Bridge, Chennai, Tamil Nadu, India. The system integrates a Python backend powered by OpenAI\'s GPT-5.4-nano model with a real-time HTML/CSS/JavaScript frontend, connected through Flask and Socket.IO. The AI receives time slot inputs, determines traffic density ranges from a lookup table based on real-world observations, and predicts realistic vehicle counts for nine lane paths: R1-R4, R1-R5, R1-R2, R6-R2, R6-R4, R6-R5, R3-R5, R3-R2, and R3-R4. Using these counts, it calculates signal timings for five units — S1, S2, S3, and pedestrian signals P1 and P2 — across five traffic cases (C-1 to C-5). Signals operate independently through Green ? Yellow ? Red phases, with transitions occurring only when all signals reach red. A midnight mode between 12:01 AM and 4:59 AM switches all signals to blinking red. The dashboard features a dark theme with LCD-style countdown timers and a manual override for emergencies. Economically viable at USD 0.20 per million tokens, the GPT-5.4-nano model demonstrates practical use of AI in structured decision-making for critical infrastructure. Results show reduced delays, improved throughput, and safer pedestrian crossings.