Building agent-native systems — from context engineering and runtime design to real-world AI agents people actually use.
I’m not just building AI demos — I’m exploring how to make agents actually usable:
- ⚙️ Tool-using agents & runtime systems
- 🧩 Reusable skills & agent infrastructure
- 🧠 Context engineering (compression, memory, token efficiency)
- 🚀 Real-world AI products (not just experiments)
Context compression engine for coding agents
- Reduces noisy CLI outputs (e.g.
git diff, logs) by 60–95% - Improves reasoning quality under limited token budgets
- Designed as a pre-processing layer for agent context
🔗 https://github.com/dethan3/ClawCompress
These are not just repos — they are products I'm actively shaping.
Language learning powered by AI agents
- Context-aware learning experience
- Personalized feedback & interaction loop
- Exploring how agents can replace traditional learning flows
Turn long content into actionable insights
- Snap articles into summaries / highlights
- Designed for fast consumption in real workflows
- Focus: information compression + usability
AI-powered tracking & awareness system
- Track what matters (signals, trends, events)
- Agent-assisted interpretation instead of raw data
- Moving from dashboards → agent insights
Making AI video (Sora-like workflows) usable
- Simplifying generation pipelines
- Bridging creativity and actual output
- Exploring agent-assisted content creation
Multi-tenant personal AI agent system
🔗 https://github.com/dethan3/Pi-Butler
- Messaging + API integrations
- Task, notes, and daily assistance
- Full agent loop implementation
Job matching agent using Claude Agent SDK
🔗 https://github.com/dethan3/jobfit
- Resume parsing + JD matching
- Automated evaluation & suggestions
Reusable skills & protocol ideas for AI agents
🔗 https://github.com/dethan3/skillshub
SKILL.mdas a standard- Cross-model skill reuse
- Early exploration of agent ecosystem layer
- Languages: TypeScript · Python
- Focus: LLM Agents · Tooling · Systems
- Infra Thinking: Runtime · Context · Abstractions
I’m less interested in one-off AI demos,
and more interested in building systems that make agents actually work.
If you're working on similar problems — feel free to reach out.