MVPeak AI
Role
Co-Developer
Team
Najaf Arash Dastnaei
Joshua Mills Lamptey
Timeline
4 Months
Overview
MVPeak was an AI-powered League of Legends coaching platform designed to help players understand and improve their gameplay. At its core was MAICRO, an AI model we trained on high-elo match data to read a player's in-game decisions the way a human coach would: spotting mistakes, weighing them against what a stronger player would have done, and explaining the reasoning in plain language. We fed MAICRO structured match events and statistical baselines from high-ranked players, and it turned that into personalised coaching output, minute-by-minute breakdowns, and benchmarks tied to a player's actual games rather than generic tips. The aim was to move beyond one-size-fits-all advice and give players insights based on what actually happened in their games.
Highlights
Our goal was to make high-level League of Legends coaching more accessible through data-driven, personalised AI analysis.
0.1Breakdown for a single match, generated by our AI model.
Image
0.2Minute-by-minute feedback tied to exact timestamps.
Image
0.3Match overview benchmarked against high-elo averages.
ImageThe Problem
Most players don't know why they are losing.
Generic advice doesn't explain what went wrong in your game.
League of Legends provides extensive match statistics, but raw numbers rarely explain the decisions that led to a loss. Players could review their games themselves, but identifying positioning mistakes, missed opportunities and better alternatives required significant time and game knowledge.
Process
Building a pipeline that could scale with the game.
From raw match data to personalised AI coaching.
We built an ETL pipeline that ingested match data from the Riot Games API, parsed game events into structured situations, stored the data in PostgreSQL, and generated statistical baselines from high-ranked players. These datasets were then combined with embeddings and an LLM to produce the final coaching reports.
A key challenge was keeping the system reliable as League of Legends changed. Champions, items and their properties are regularly updated, so we built a dedicated pipeline interface that allowed us to update and manage this data without modifying the underlying code. This made adapting to new patches significantly faster and reduced the need for manual code changes across the system.
Solution
An end-to-end coaching pipeline built for scale and change.
From raw Riot data to actionable, personalised coaching.
The final system tied match ingestion, game-state analysis, statistical baselines, and AI coaching into a single pipeline, with expensive processing handled offline so live requests could return a player report quickly while the AI coaching layer ran separately. The management interface described above did double duty here too: it's what kept the whole pipeline maintainable as League itself kept changing, patch after patch.
Outcome
From prototype to a working product to our decision to drop the project.
Cutting our losses before it was too late.
MVPeak progressed from an early concept into a functional AI coaching platform for League of Legends. The system processed real match data, compared player decisions against high-elo gameplay, and generated personalised feedback. Building the full pipeline gave us a strong technical foundation and proved that the concept was viable. However, maintaining a project of this scale required infrastructure and API costs that were beyond our budget. Rather than continuing to spend money on a product we could not realistically sustain, we made the decision to stop development and cut our losses.
