AI-Powered Cricket Management Simulation Game App

Creating a realistic sports management simulator presents unique technical challenges: managing strict, probabilistic math for match outcomes, handling highly concurrent multiplayer auction economies, and keeping textual commentary fresh and engaging across thousands of simulated matches. I engineered the backend for this cricket club manager, a comprehensive NestJS and Supabase-based application that seamlessly blends a highly complex sports simulation engine with the generative capabilities of Google's Gemini AI to deliver a hyper-realistic, endlessly varied gaming experience.

I built a robust backend architecture using NestJS and Supabase PostgreSQL. The system integrates advanced AI specifically designed to hydrate the game's strict, mathematical simulation events with natural language. To ensure the game felt alive, I engineered an AI-driven commentary template generation pipeline, where the LLM contextualizes the exact state of the match (e.g., pitch conditions, bowler type, batter aggression, and shot control) into dynamic prose. Additionally, I built an automated, globally synchronized virtual economy engine that manages daily player generation, squad contracts, and live bidding wars, utilizing secure database transactions to handle high-frequency auction concurrency.

Deterministic Match Engine: A complex state-machine that evaluates bowler styles, pitch deterioration, batter stamina, running aggression, and real-time fielding presets to mathematically calculate the outcome of every single delivery across T10, T20, and ODI cricket match formats.

AI-Generated Ball-by-Ball Commentary: A scheduled serverless pipeline powered by Gemini AI that dynamically generates contextual commentary templates based on highly specific ball outcomes (e.g., a mistimed pull shot off a fast bowler on a green pitch), completely eliminating repetitive game text.

Live Multiplayer Player Auctions: A fully automated transfer market featuring dynamic player value algorithms, anti-snipe bid extensions, and scheduled server-side cron jobs that backfill global player supply daily.

Simulated Decision Review System (DRS): A mathematically rigorous recreation of real-world cricket technology (Hawk-Eye and Ultra-Edge). It calculates ball pitching zones, impact alignment, and exact wicket-hitting probabilities to allow managers to challenge umpire decisions.

The Issue: Traditional sports games rely on a static, hardcoded database of commentary lines. After a few hours of gameplay, users hear the exact same phrases repeated endlessly, which destroys immersion, especially in a simulator where thousands of balls are bowled per day.

The Solution: I implemented an AI-driven content generation pipeline. Instead of writing thousands of lines of dialogue manually, the backend continually queries Gemini AI with extremely specific simulation constraints (e.g., "Write a commentary line for a Right Arm Fast bowler inducing a thick outside edge that drops short of first slip").

The Issue: Handling virtual currency bids on live player auctions requires bulletproof transaction logic. If multiple users bid on a highly sought-after player in the final seconds of an auction, race conditions could lead to conflicting database states, negative balances, or lost virtual currency.

The Solution: I engineered a highly secure bidding pipeline using atomic transactions. When a user submits a bid, the backend first validates their virtual currency balance. It then executes a strict update sequence that locks the auction row, verifies the new bid is higher than the current_bid plus the BID_INCREMENT, deducts the currency from the new bidder, and refunds the previous highest_bidder_id simultaneously.

The Issue: Cricket is incredibly complex. Simulating a match requires tracking the physical stamina of fast bowlers vs. spinners, incorporating superiority of one team's ratings over the opponent, the psychological mood of the squad, the exact placement of 11 fielders on a coordinate grid, and the deterioration of the pitch over 300 deliveries.

The Solution: I engineered a deeply layered algorithm that calculates a base probability matrix for every ball. The system modifies these probabilities in real-time based on live manager inputs (e.g., changing batter aggression or moving a fielder to "deep third man").