available component
LLM Gateway Starter
An OpenAI-compatible model gateway with routing, fallback, rate limits, usage accounting, and Docker Compose deployment.
Blog
Filtering new models, frameworks, and tools through the lens of production engineering.
Making Agents Handle Follow-ups for Real
A practical architecture note on why agent follow-ups should not be treated as plain chat, and how session, run, memory, DAG versioning, and runtime events fit together.
Refactoring an Agent Runtime from PCA to ReAct
A practical review of why static DAG planning became heavy for general-purpose tasks, and what runtime constraints ReAct needed to work safely.
Fine-Tuning a 7B Model for a Business Task
A sanitized enterprise-assessment example showing what 7B fine-tuning can and cannot solve, and how to run LoRA/QLoRA with LLaMA-Factory.
Testing 5 Open-Source PDF Parsing Tools
A reproducible comparison of Docling, MinerU, PaddleOCR-VL-1.6, DeepSeek-OCR 2, and dots.mocr on OmniDocBench article-300 using the same test server.
Notes
Short notes, observations, and image records.
Projects
Reusable starter kits, architecture templates, and AI engineering projects built from production experience.
available component
LLM Observability Starter
A self-hosted LLM observability collector for traces, latency, token usage, cost, feedback, and eval events.