Jingxiao Cai, Ph.D.

Current focus — Enterprise AI Metadata Retrieval (Feb 2026 – Present)

• Contribute to backend services and agent-facing retrieval tools that map natural-language intent to relevant schema, table, and column metadata for downstream analytics and text-to-SQL workflows.
• Develop and validate service APIs, retrieval workflows, contract tests, and end-to-end integrations spanning metadata annotation, semantic search, and schema discovery.
• Evaluate metadata-retrieval approaches using public text-to-SQL benchmarks.

MySQL HeatWave AutoML/GenAI — Distributed ML Runtime Systems (Oct 2018 – Jan 2026)

• Built and hardened the Dask-based backend runtime behind HeatWave ML, spanning SQL-to-backend-runtime execution paths, secure transport, worker lifecycle, and service reliability as the product evolved into a production cloud service.
• Implemented and hardened secure communication and trust paths across the ML runtime, including database-to-ML-driver transport, encrypted Dask worker communication, worker-to-worker mTLS, and certificate hardening.
• Hardened runtime control paths for concurrency, process isolation, abort/recovery, cluster health, and elastic worker management, including readiness work at 100+ node scale.
• Drove root-cause analysis (RCA) and reliability improvements for hard runtime failures, OOM/hybrid workload contention, and distributed execution issues across backend/runtime paths.

• Built talent recommendation engine using Random Forest to optimize hiring for Store/Assistant Managers, reducing vacancy impact.
• Created internal wiki on data science techniques (RegEx, imputation, Linux scripting).

• Developed ML algorithms (ANN/Random Forest) for radar turbulence detection in collaboration with Hong Kong Observatory.
• Optimized GPU-parallelized radar algorithms (Super-Resolution, Pulse Compression) on Nvidia Jetson.
• Implemented target tracking for wind turbine clutters.

• Contributed 13 merged upstream PRs to OpenClaw with regression coverage, focused on AI-agent gateway/runtime reliability, tool-progress delivery, and document extraction correctness; selected PRs: #98021 (GPT-5.6 Ultra cross-runtime integration), #88159 (logs-follow journal fallback retry), #80947 (QMD session-recall gate diagnostics), #84708 (message-tool mirror replay recovery), #89279 (Discord ACP thread completion delivery), #90487 (ChatGPT/Codex Responses SSE stream hardening), and 7 more.
• Listed as @anyech on the official contributors page; work includes TypeScript gateway/runtime correctness, Discord/tool-progress delivery, PDF.js/package-layout handling, and LLM-integrated document extraction.