AI Agent Developer | Autonomous Systems | Developer Tools
Building intelligent solutions that enhance developer productivity
I'm passionate about creating AI-powered autonomous agents and intelligent developer tools that transform how teams write, review, and manage code. My work focuses on bridging the gap between cutting-edge AI technology and practical developer needs.
1. π€ Mizumi - Self-Learning PR Review Agent
GitHub Action that reviews pull requests using AI, learns from past reviews, and posts actionable findings β with deterministic rules that never hallucinate.
Key Features:
- β BYOK from day 1 (Anthropic, OpenAI, Google, NVIDIA NIM, OpenRouter, or any OpenAI-compatible endpoint)
- β Self-learning via repository memory
- β Deterministic rules (catches hardcoded secrets, missing auth, SQL injection WITHOUT LLM calls)
- β Two-pass review (LLM review + self-critique on cheaper model)
- β Prompt injection defense framework (first AI code review tool with explicit defense architecture)
- β Confidence calibration & auto-fix on π reaction
- β CI-validated fix loop (apply β poll β revert on failure)
- β AST cross-file contract analysis
2. π§ Hermes Rate Limiter Plugin - Smart Rate Limiting
Cross-session rate limit guard for Hermes Agent side clients that prevents retry amplification when rate limits are hit.
Problem Solved: Each 429 error triggers up to 9 API calls (3 SDK retries Γ 3 Hermes retries). This plugin records rate limit state on first 429 and checks it before subsequent attempts across all sessions.
Key Features:
- β Cross-session tracking with shared state file
- β Retry amplification prevention
- β
Runtime controls via slash commands (
/ratelimit status,/ratelimit clear, etc.) - β Thread-safe atomic writes
- β Reset time parsing (x-ratelimit headers, retry-after, default cooldown)
3. πΎ Noxem - AI Memory System for Hermes
Two-brain AI memory system with hybrid semantic search (vector KNN + FTS5) and optional advisor brain via QwenProxy or local LLM.
Brain 1 β Semantic Engine:
- Hybrid search with Reciprocal Rank Fusion
- Auto-categorization (12+ tags)
- Smart deduplication (cosine >0.92)
- Conflict resolution & contextual enrichment
- Spaced repetition & importance scoring
- Provenance graph tracking
Brain 2 β Reasoning Engine (optional):
- Cloud (QwenProxy) or local (Ollama, LM Studio, llama.cpp)
- Drift detection & context recovery
- Background web research via DuckDuckGo
- Session extraction & consolidation
- OpenAI-compatible API on port 8000
- π Exploring: Advanced AI agent architectures and intelligent memory systems
- π± Learning: Defense mechanisms, prompt injection mitigation, and security best practices
- π‘ Building: Next-generation autonomous development tools and AI agents
- π€ Contributing: To open-source AI ecosystems
I'm always excited to collaborate on:
- π€ AI agent development and memory systems
- π οΈ Developer tools and productivity solutions
- π AI safety, security, and defense mechanisms
- π Open-source projects
Feel free to:
- β Star projects you find interesting
- π Report issues and suggest improvements
- π€ Submit pull requests
- π Reach out with ideas and feedback
Let's build the future of AI-powered development together! π