Independent Developer — Agentic AI, Software Engineering & Quantitative Systems
Agronomy Major · Computer Science Minor
Nanjing Agricultural University
About · Areas of interest · Selected work · Toolkit · Activity
Academic training at Nanjing Agricultural University combines a major in Agronomy with a formally admitted minor in Computer Science. Beyond the classroom, independent software development provides a practical foundation for exploring intelligent systems, research-oriented applications, and the engineering challenges that emerge at the intersection of disciplines.
Particular attention goes to the engineering of agentic AI: the architecture of LLM-powered systems and the mechanisms through which language models interact with external tools, coordinate multi-stage workflows, and operate within well-defined constraints. Context management, orchestration, evaluation, and human oversight matter here because they distinguish a useful autonomous workflow from a convincing demonstration.
Alongside applied AI, quantitative research infrastructure forms another substantial focus. Projects including Alpha Mosaic and Meridian Alpha explore financial data organization, analytical workflows, strategy evaluation, and paper-only decision-support systems. The emphasis rests on methodological transparency, reproducibility, and the integration of computational research with informed human judgment.
Python constitutes the primary development language, complemented by C/C++ and JavaScript/TypeScript. Current technical interests also extend toward software architecture, developer tooling, AI-assisted engineering workflows, and the integration of intelligent software with real-world systems — particularly automation, smart agriculture, and intelligent manufacturing.
Underlying these pursuits is a consistent engineering principle: functionality alone is insufficient without an understanding of the assumptions, limitations, and trade-offs that shape a system's behavior. The objective is not simply to make software work, but to make its design intelligible, its outputs verifiable, and its evolution sustainable.
Agentic systems
Tool integration · Workflow orchestration · Context management
Software engineering
Developer tooling · Automation · Software architecture
Quantitative systems
Research infrastructure · Data pipelines · Strategy evaluation
Applied intelligence
Intelligent systems · Smart agriculture · Industrial automation
Document engineering for everyday work. CleanText Studio turns copied or AI-assisted text into editable documents while retaining useful structure. A shared document tree powers preview and DOCX export; basic cleanup runs locally, with optional AI refinement initiated by the user.
Python · structure-aware parsing · DOCX/TXT export · local-first processing
Source · Latest release
An evidence-first quantitative research workbench. It discovers a bounded US equity universe, validates available market data, computes deterministic factors, and saves research reports. Provider health, evidence, factor calculations, and report generation remain explicit; historical replay is diagnostic and paper evaluation stays separate from live trading. The project makes no claim of demonstrated alpha.
Python · deterministic factors · evidence-gated evaluation · read-only MCP
Source · Architecture
Decision support with explicit gates. Meridian connects research evidence to deterministic risk checks, status-aware reports, and a persistent paper ledger. Its stateful pipeline separates data, research, decision, risk, and paper stages. It is an independent project from Alpha Mosaic. Its current boundary is PAPER_ONLY: broker orders and real-account execution are disabled.
Python · stateful workflows · risk gates · paper ledger
Source · System boundaries
RecallForge AI — A full-stack study workspace that turns course materials into source-anchored review, active recall, and exportable study packs. Generated material remains a study aid to verify against its sources. FastAPI · React · v0.6.0 release
RecallForge Skill — A host-executed Agent Skill for inspecting learning materials, reconstructing course knowledge, and testing recall. Visual understanding depends on the selected host's capabilities. Agent Skills · Python · v2.2.0 release
Languages — Python · C · C++ · JavaScript · TypeScript
Development — Git · GitHub · GitHub Actions · FastAPI · React · Docker · SQLite
AI & research — LLM applications · agent workflows · tool integration · research automation · data processing · quantitative research
Contribution trail
Activity assets are regenerated through this repository's GitHub Actions. The contribution trail uses snk and stays secondary to the project notes above.