AIWF is Shawn O'Hagan's public engineering lane for local-first creative tools, model training interfaces, model operating systems, source-backed guides, and consumer-hardware experiments. The repositories show the work; the services page turns a focused part of it into a client scope.
Geoffrey Hinton explains the ideas behind current AI systems in a clear, accessible conversation. It is an external educational reference, not an AIWF claim or company endorsement.
External video: “Is AI Hiding Its Full Power? With Geoffrey Hinton.” Watch on YouTube.
Each card links to the public source. Maturity varies by repository, so active development and stable sharing are kept separate.
Local-first image, inpaint, model browsing, generation, and video workflow development for Windows and consumer GPUs.
Open AIWF StudioA training UI for consumer hardware with explicit run configuration, receipts, model preparation, and local workflow goals.
Open ReTrainResearch into memory-efficient model routing, specialized components, and a practical operating layer for smaller systems.
Open Model Operating KernelReusable Codex workflows for research, UI review, security boundaries, source handling, and local AI engineering.
Open the skill packPractical local-AI guidance for setup, tools, failure modes, hardware expectations, and work that should stay reviewable.
Open the guidesTests whether a small adapter can follow source rules, retrieval labels, and refusal behavior without pretending the corpus is memorized truth.
Open Atlas LoRAAIWF provides public proof. Client work stays private, scoped, and tied to the customer's approved files and systems.
We map the task, data, risk, and expected gain, then recommend AI, ordinary automation, or a simpler rule-based system.
One narrow assistant or automation, one approved source lane, a human review step, and a small acceptance test.
A controlled internal source for company files, procedures, project notes, and approved question-answer work.
Dataset condition, licensing, base-model fit, compute, evaluation, and the difference between fine-tuning and retrieval.
One approved training scope with configuration, run receipts, evaluation, and artifact handoff after readiness passes.
Company-specific training on model choice, prompting, verification, privacy, file handling, and token control.
Private files and latency-sensitive work can stay on controlled hardware when the model, storage, and operating budget support it.
Managed APIs can be the right choice for speed, capability, or a small starting budget. Provider costs and data handling are stated before use.
When the job is to answer from changing company files, grounded retrieval may be cheaper and easier to update than model tuning.
A demo response is not a result. We define test prompts, expected behavior, failure handling, and the record needed to compare changes.
Daily development should not break the company download. Public files will point to a stable branch or versioned release artifact after a smoke gate.
Use the services intake for a private workflow, company notebook, model review, fine-tune, or employee workshop.
Use the investor route for AIWF sponsorship, technical diligence, strategic backing, or a public-interest collaboration.