Academia × Engineering Dual Engine
Xianma Intelligent Technology was co-founded by academic authorities and senior engineering teams, dedicated to deeply integrating LLMs, multi-agent systems, and domain expertise to build truly deployable vertical AI products.
Company Overview
Company: Xiamen Xianma Intelligent Technology Co., Ltd.
Website: www.xianma.top
Positioning: Vertical AI Agent Products & Solutions Provider
Location: Xiamen, Fujian (Wuhan OPC planned)
Core Fields: Government, Finance, Manufacturing, Consumer, Content
Experience: Founder's 27 years of full-stack & AI architecture experience
Academic Backing: Backed by academic authorities from research institutes
Product Matrix: 15 active / live projects
Tech Stack: LLM Fine-tuning · Multi-Agent · Knowledge Graph
Theory Meets Practice
Academic authority + senior engineering, deeply integrated
Xiong Chao
Founder & Tech Lead, Xianma Intelligent Tech
Former CTO of a NEEQ-listed enterprise ("Xiamen 258 Group")
Education
• M.S. in Computer Applied Tech, Peking University
• Thesis: Controllable Text Generation Methods
• B.Eng., Wuhan University of Technology
Expertise
• LLM Fine-tuning & Deployment
• Big-data Systems
• Multi-Agent
• Knowledge Graph
• Full-stack Dev
• Cloud-native Architecture
Certifications
• Microsoft BizTalk Senior Instructor
• Neo4j Official Certification
• Lead of algorithm-safety filing for Woqi Digital Avatar
Prof. Zhang Yang
Director, Institute of New Economy & Finance, Shenzhen University
FinTech & Regulation Authority · PhD Supervisor
Education
• Postdoc in Economics, Peking University
• Ph.D. in Management, Wuhan University
• Visiting Scholar, Johns Hopkins University
Key Monograph
Research on International Experience and Chinese Practice in FinTech Regulation
Economic Science Press, 2025
Honors & Awards
• Dong Fureng Economics Award, Wuhan University
• Most Influential Scholar in China (2020)
• Shenzhen High-level Talent
Synergy
Academic Authority
Prof. Zhang's monograph provides theoretical foundation
Technical Prowess
27 years of experience ensures system reliability
Industry-Academia
Deep integration of theory and practice
Xiong Chao · Career Timeline
From engineering roots to vertical AI agents — 27 years of evolution
B.Eng., Wuhan University of Technology
Wuhan University of Technology
Engineering foundation; start of a 27-year technical career.
Enterprise Integration & Full-Stack
Microsoft BizTalk Senior Instructor
Deep in EAI, BizTalk and full-stack engineering; certified Microsoft BizTalk Senior Instructor.
M.S. in Computer Applied Tech, Peking University
Peking University
Thesis "Controllable Text Generation Methods" — foundation in NLP & generative AI.
CTO, Xiamen 258 Group
NEEQ-listed enterprise
Led architecture and teams for large-scale internet platforms.
Invention Patents × 3: Fructose Dispenser / Mobile Remote Control
CNIPA
3 invention patents filed in the same year: ① auto-controlled fructose dispenser (scan + Bluetooth); ② Android APP remote control system (AccessibilityService + TCP long-connection); ③ web-based mobile device control (scripted group-control architecture).
Journal Paper: Improving Text Generation via Sampling Suppression
Information & Computer: Theory Ed., Vol.10, 2021
Proposed suppressing specific token sampling probabilities during autoregressive decoding to mitigate repetition/hallucination; successfully applied in multiple projects.
Master's Thesis: Controllable Text Generation via Prompt Learning
Peking Univ. Shenzhen · Advisor Prof. Zou Yuexian
Focused on Prompt Learning and controllable text generation with system implementation and experimental validation.
Algorithm-Safety Filing Lead, Woqi Digital Avatar
Woqi Digital Avatar
Led generative-AI digital-human algorithm safety compliance filing.
Founded Xiamen Xianma Intelligent Tech
Founder & Tech Lead
Focused on vertical AI agents fusing LLMs, multi-agent systems and domain knowledge.
AMD AI Certification Track (5)
AMD
Completed AI On AMD / R1 Fine-tune / K8s on Instinct / Multi-Agent systems.
Academia × Engineering Dual Engine
Prof. Zhang Yang, Shenzhen Univ.
Industry-academia synergy with a FinTech regulation authority.
Some years are inferred from public résumé; corrections welcome.
Certifications & Invention Patents
Authoritative certifications + own IP, building a trustworthy AI foundation
AMD AI Certification Track
Completed AMD full-stack AI certifications in 2025–2026 (click to view original certificates)
AMD AI Cert 1 · AI Fundamentals
AMD · 2025–2026
AMD AI Cert 2 · AI On AMD
AMD · 2025–2026
AMD AI Cert 3 · Fine-tune R1 on Unsloth
AMD · 2025–2026
AMD AI Cert 4 · K8s on AMD Instinct GPUs
AMD · 2025–2026
AMD AI Cert 5 · AI Agents 201 Multi-Agent
AMD · 2025–2026
Professional Credentials
Amateur Radio
Licensed Ham Radio Operator
BH6TKJ
Class A Amateur Radio Station
Algorithm Filing
Internet Information Service Algorithm Filing · CAC Regulatory Compliance
Woqi Digital Avatar Algorithm
Suzhou Woqi Artificial Intelligence Technology Co., Ltd.
Invention Patents
Published own intellectual property
A Fructose-Dispensing Control System with Automatic Sugar Output and Its Control Method
CN107203796A · Published 2017-09-26 · Inventors: Xiong Chao, Zhang Jingjie
Abstract
Aiming at the problems of physical-button dependence and closed, non-upgradable systems in existing fructose dispensers, the invention proposes automatic dispensing via barcode recognition and wireless communication: a POS prints a barcode label, a scanner decodes the product ID and uploads it wirelessly to a host, which then issues a wireless dispense command. A Bluetooth slave module externally attached to the control board turns the closed system into an open, externally controllable one, supporting status feedback and centralized multi-machine management, greatly reducing training and misoperation cost.
Fig.1 · Flow Diagram
POS prints barcode → scan & decode → host issues command → fructose machine dispenses → status feedback → POS alerts
System and Method for Remote Control of Android Apps
CN107483576B · Granted 2020-09-18 · Inventors: Xiong Chao, Hu Xiaoting, Wen Zongxuan
Abstract
This invention discloses a system and method for remote control of Android apps, comprising a web client, a master server, and mobile devices with an auxiliary-function (AccessibilityService) app. A TCP/IP long connection is established between the server and the auxiliary app. The web client sends binding authorization and control instructions (JavaScript scripts) via the server, which are executed by the AccessibilityService to control other target apps on the device. This enables web-based remote group control without ROOT or physical connections, applicable to telesales automation and multi-device cluster testing.
System and Method for Controlling Mobile Devices via Web Client
CN107317882A · Published 2017-11-03 · Inventors: Xiong Chao, Zhuang Liangji, Hu Xiaoting, Wen Zongxuan · Applicant: Xiamen Domino Data Technology Co Ltd
Abstract
This invention discloses a system and method for controlling mobile devices via a web client, comprising a web frontend, a master server, and mobile devices with an authorized app. A TCP/IP long connection is established between the server and mobile app. The web client sends binding authorization or control requests (scripted commands), and the mobile app automatically executes the specified functions (phone calls, SMS, app launch) and returns execution results. The system supports bidirectional notification and control — incoming phone calls can pop up on PC with answer/reject options, enabling seamless PC-mobile collaboration.
Academic Papers & Research
Journal Articles · Thesis · Core Algorithm Research
Improving Text Generation Quality by Suppressing Sampling Probability
Information & Computer: Theory Edition · 2021 Vol.10 | pp.12–15 | 4 pages
Abstract
With the optimization of transformer encoder-decoder architectures and large-scale unsupervised text data training, pre-trained language models like GPT-2 have become increasingly powerful. Generating text via autoregressive next-token prediction essentially samples from the probability distribution conditioned on prior context. Without proper control, generated text easily falls into repetitive loops or produces hallucinations — internal contradictions and factual errors. This paper proposes suppressing specific token sampling probabilities during decoding, achieving higher-quality, more fluent, and longer text passages in open-ended continuation tasks. The method applies to various generative pre-trained language models without additional training or fine-tuning — only targeted suppression token sets are needed, making it a practical, low-computation-cost approach.
Research and System Implementation of Controllable Text Generation Based on Prompt Learning
Master's Degree Thesis (Equivalent) · June 2022
Abstract
Research direction: AI & NLP. Focused on Prompt Learning applications in controllable text generation, completing both theoretical research and system prototype implementation.
Build Industry AI with Us
Explore our product matrix, or contact the team directly.