Commercial delivery: connected match data, supplemental research, analysis reports and paid use; launched with paying users and is now offline.
AI product / Data pipeline
Football AI · Autonomous analysis and a paid product
Applied AI across development and service operation: automated collection and cleaning, research, predictive analysis, self-checks and result synchronization. The product launched and served paying customers.
WORK & CONTRIBUTIONS
My work on this project.
Runtime architecture: one codebase with an online business node and a data-AI node, separating user/task APIs from collection, cleaning, research, generation and synchronization.
Agent orchestration: stage-specific skills for research planning, extraction and report generation; research can search, while drafting reads the prepared input.
Self-checking: skill hashes, output schemas and references combine with task locks, heartbeats, timeouts and failure states for automated processing and diagnosis.
Multi-AI cost design: Codex/Claude development collaboration, Claude Code agent execution and DeepSeek gateway adaptation, with context compaction, caching and turn limits to control cost.
NEXT PROJECT