Building AI Without Fear — practical local AI tools, RAG systems, and ComfyUI workflows for people who want AI to actually work.
AI Without Fear Atlas — A grounded RAG corpus for local AI workflows. Covers ComfyUI nodes and API, Gradio 6, model serving, Python/pip troubleshooting, and evaluation harnesses. Built so AI assistants can retrieve real answers instead of hallucinating setup steps.
The problem it solves: most AI assistants confidently get ComfyUI node names wrong, recommend outdated Gradio patterns, and hallucinate venv paths. This corpus gives them grounded retrieval material instead.
Atlas Reader LoRA Lab — A working research/evaluation lab for testing whether a lightweight QLoRA adapter can learn to read structured Atlas context: lanes, cards, source rules, compact evidence packs, and off-ramp behavior when evidence is missing. The current best internal run preserves compact-card performance, improves targeted retrieval behavior, and records compact selected-card paths using about 5.05x-20x fewer total tokens than raw workspace/RAG-style comparisons in specific lab use cases. It is not presented as a production package, external benchmark, or universal token-reduction claim.
- Local AI — Running and fine-tuning models on consumer hardware (16GB / 8GB VRAM)
- ComfyUI — Workflows, node ecosystem, API automation, datatype boundaries
- RAG & Retrieval — Corpus design, chunking strategy, source governance, answer gating
- Gradio 6 — App building, production deployment, ComfyUI bridge integration
- Model Training — LoRA/QLoRA experiments, evaluation records, and domain-specific adapters
If the Atlas or any of my tools have saved you time, you can support continued development:
AI Without Fear — practical local AI for real people, on real hardware.