Erikdai
Erikdai is the personal research and open-source portfolio of Chengxiao Dai, an AI researcher, agent engineer and University of Sydney computer-science master's student. Dai works on large-language-model post-training, reinforcement learning for agentic systems, knowledge-graph reasoning and domain applications, while publishing software and research openly.
At a Glance
- AI/ML researchers and graduate students
- Developers building LLM agents and agentic memory systems
- Scientific-computing and AI-for-science communities
- Quantitative-investment researchers interested in A-share paper trading
- +2 more
AI Tools by Erikdai
(1)
ShardMemo
Agentic Memory Service for LLMs
Discussions
No discussions yet
Be the first to start a discussion about Erikdai
Latest News
ShardMemo repository made its initial public commit, presenting scope-before-routing memory retrieval for agentic systems.
QuantCouncil repository work renamed the project and added user holdings plus PDF/Word report export around the multi-agent paper-trading workflow.
BridgeAgent alpha release added deterministic DOM/UIA routing, LLM-free skill replay, guard rails and CLI/GUI/MCP entry points.
Dai launched a public notes/blog page, beginning with a post about keeping research notes public.
Products & Services
An MIT-licensed three-tier memory service for LLM agents. It keeps bounded active context, stores evidence in shards with shard-local approximate-nearest-neighbor indexes, and maintains a versioned skill library; metadata predicates enforce scope before a learned router probes eligible shards.
An A-share multi-agent research and paper-trading platform built on TradingAgents, with analyst agents, bull/bear debate, a trader and risk committee, a rule-based risk gate, human approval and a paper ledger. It uses LangGraph, MCP, RAG, FastAPI and React.
A Windows AI agent that unifies browser DOM access and native Windows UI Automation in one workspace, with deterministic focus-based routing, LLM-free skill replay, permission guard rails and CLI, GUI and MCP entry points.
An open-access ecosystem for scientific discovery generated by AI scientists. Its multi-agent architecture supports proposal/paper submission, review and iterative refinement by human and AI scientists, with API and MCP interfaces.
Market Position
Erikdai is best understood as an individual research-and-build portfolio, not a conventional company competing through sales or headcount. Its open-source positioning emphasizes research prototypes that combine agent orchestration with explicit controls: ShardMemo separates hard access scope from learned relevance routing, BridgeAgent emphasizes deterministic DOM/UIA routing and replay, and QuantCouncil combines multi-agent analysis with a risk gate, human approval and paper trading. This differentiates the portfolio from less control-oriented generic agent demos while remaining research/open-source rather than a packaged enterprise vendor.
Leadership
Founders
Chengxiao Dai
University of Sydney Master of Computer Science (Advanced Entry) student and tutor; previously a research intern at A*STAR's Singapore Institute of Manufacturing Technology and the China Academy of Information and Communications Technology, and an algorithm engineer at Fujian Imperial Vision Technology Group. Dai is also identified as a co-founder of ChuChiang Data on the personal site.
Executive Team
Chengxiao Dai
Independent AI researcher, agent engineer and open-source creator
Master of Computer Science (Advanced Entry) student and tutor at the University of Sydney; former A*STAR SIMTech and CAICT research intern and Imperial Vision algorithm engineer.
Business Model
Revenue Model
The subject is presented as an individual researcher and open-source creator rather than a commercial company. The listed software is publicly available, including MIT-licensed repositories; no paid subscription, API pricing or commercial revenue model was identified.
Target Markets
- AI/ML researchers and graduate students
- Developers building LLM agents and agentic memory systems
- Scientific-computing and AI-for-science communities
- Quantitative-investment researchers interested in A-share paper trading
- Automation developers targeting Windows and browser workflows
- Knowledge-graph, legal-tech and domain-specific question-answering researchers
- Persistent memory retrieval for LLM and tool-using agents
- Conversational, long-context and procedural-memory benchmarks such as LoCoMo, HotpotQA and ToolBench
- A-share investment research and paper trading
- Browser and Windows desktop automation
- AI-generated scientific proposal, paper review and publication workflows
- Discovery and reuse of agent skills, prompts, MCP servers and domain resources