Xiong Xiao

CATARC · Embedded Software Development Intern

Worked on an IoT communication control module based on STM32WLE, covering multi-interface driver abstraction, structured storage deployment, and high-precision clock monitoring.

• Decoupled HAL/BSP and application logic, built layered task management, and improved code reuse by more than 50%.
• Ported FatFS for persistent sensor-data storage and used RTC handling to improve timestamp stability.
• Developed the GX30H05 multi-channel acquisition driver over a one-wire path, using logic analyzer and oscilloscope traces to locate communication errors caused by interface level mismatch.
• Debugging documents and compatibility recommendations were incorporated into team interface standards.

CATARC Multi-Interface Embedded Development and Debugging

Organized STM32WLE interface drivers, FatFS storage, RTC timestamps, and one-wire acquisition debugging into a HAL/BSP/App layered structure, improving code reuse by more than 50%.

Smart Life Emotional Companion System

Split AI interaction, environment sampling, and display feedback with threads so network requests do not block local sampling and safety-alert paths.

3D Machine Vision Chinese Chess Robot

Connected YOLOv5 recognition, coordinate mapping, Alpha-Beta decision, and PWM robotic-arm control, reaching +/-0.5 mm repeat positioning and response within 300 ms.

Data Knowledge Insight Demo

Built an end-to-end “data to knowledge to insight” agent demo over medical sample data, validating an engineering loop from raw data to structured knowledge, question answering, and analysis reports.

• Task Loop: implemented pipelines for data processing, knowledge graph construction, KGQA, and statistical analysis, runnable with one command through run_demo.py.
• Engineering: rule-based planning with LLM planning fallback; FastAPI exposes five endpoints across health check and three tasks; 56 pytest tests pass.
• Platform Fit: designed DataMate-style mappers and a Nexent/FastMCP-style MCP server with four tools callable from agent platforms.
• Open Source: won Third Prize in the ModelEngine Open Source Project Contribution Competition (2026.08), with code merged into the community AgentsHub repository.
• Outputs: cleaned records, a medical knowledge graph (JSON/GraphML/HTML), analysis reports, charts, and a SQLite database.

AI Invoice Batch Organizer and Price Review Tool

Built an Electron + React invoice tool with OCR recognition, batch organization, local price evidence, and LLM review suggestions, while keeping final rename actions manually confirmed.