Open-source AI systems and research software
Grounded Agents · Human-in-the-Loop APIs · Automation · Computer Vision · Scientific Modeling
| Destination | Description |
|---|---|
| jz-ai-chat.com | Live portfolio, project showcase, and interactive agent |
| All repositories | Complete public project catalog |
I am a student at Princeton International School of Mathematics and Science.
I build open-source systems that turn ambiguous real-world problems into transparent, testable software: typed schemas, retrieval pipelines, human-in-the-loop workflows, validators, simulations, and automation with safe defaults.
My main interests are:
- Grounded AI agents that answer from inspectable, versioned knowledge
- Human-in-the-loop systems with atomic ownership, audit trails, and protocol-safe outputs
- Safe automation with OCR checks, dry runs, status indicators, and audit logs
- Computer vision for screen recognition and decision support
- Research software for biology, environmental data, simulation, and modeling
- Music technology for sheet-music conversion, transposition, and workflow automation
- Game / CLI tools that turn algorithms into interactive systems
This profile highlights a small set of active, public repositories. The complete catalog remains available on the Repositories tab.
| Project | Focus | Stack |
|---|---|---|
| Agent-Me | Grounded personal Q&A framework with local retrieval and multilingual UI. | Python FastAPI React |
| Human API | OpenAI-compatible human-response queue with transactional coordination. | PostgreSQL FastAPI React |
| IntentForge | Deterministic CAD generation with explicit intent schemas and validation. | Python CadQuery MCP |
| dot-studio | Local-first console for composing and delivering content to connected displays. | TypeScript Next.js Canvas |
| pcr532-development-platform | Browser-based NFC diagnostics, reading, and controlled card-writing workflows. | TypeScript Web Serial NFC |
| wechat-assistant | Guarded macOS automation with OCR verification, dry runs, and audit logs. | Python OCR Tkinter |
| quehun | Screen recognition and decision support for Mahjong hand analysis. | Python OpenCV Computer Vision |
| PlanetPulse | Interactive environmental monitoring with live NASA data and explainable ranking. | TypeScript Next.js Three.js |
| algae-research-toolkit | Consolidated algae simulation, PPO experiments, exposure logging, and analysis. | Python Gymnasium Research |
| music-score-toolkit | Testable MuseScore transposition and explicit score-conversion workflows. | Python MSCZ MusicXML |
| PDFtoMSCZ | PDF-to-editable-score workflow using Audiveris OMR and MuseScore 4. | Python OMR MuseScore |
| internet-meme-radar-skill | Evidence-based research workflow for fast-moving internet culture. | Python Research Codex |
Natural language prompt
↓
Structured intent schema
↓
Validation rules
↓
Generated artifact
↓
Tests, benchmarks, and real use
I am especially interested in domain-specific and human-in-the-loop agents: systems that ground answers in visible evidence, reject invalid inputs, request missing constraints, produce structured outputs, coordinate people safely, and verify results.
Languages
Python · Java · C++ · JavaScript · HTML/CSS
AI / Automation
OCR · Selenium · FastAPI · Tkinter · OpenCV · agent tooling · workflow design
Research / Modeling
Simulation · reinforcement learning · data visualization · scientific measurement · validation rules
Engineering Habits
Clear schemas · safe defaults · reproducible examples · readable docs · testing and benchmarks
Start from a real problem.
Turn it into a structured model.
Build the smallest useful system.
Measure where it fails.
Add validation, safety, and feedback loops.
Repeat until the tool becomes actually useful.
- Website: jz-ai-chat.com
- Email: [email protected]
- GitHub: @jzjzzzzzzz