AI-Trainer-OBLIVION

The arc

This is the sequel.

AI-Trainer-MAX teaches a Windows user with no prior experience how to install, run, and own a local AI — all the way through to building a digital legacy their family can still run after they're gone.

OBLIVION is what comes after that.

It is the research-math layer MAX deliberately left out. The part where the training wheels come off and you stop being a user of models and start being someone who shapes them.

It is not for beginners. It is not hand-holding.
It assumes you finished MAX and want more.

Planned arc

What OBLIVION will cover.

No release date yet. This is the terrain. Each item maps to a module or cluster of modules once the curriculum is finalized.

Fine-tuning on your own hardware

The actual mechanics. Not hand-waving. Not someone else's Colab.

LoRA / QLoRA on consumer GPUs

8GB, 12GB, 24GB — what's possible at each tier and the honest trade-offs.

Quantization trade-offs

GGUF, Q4 vs Q5 vs Q8, when each wins, when they break.

Custom Modelfiles at depth

System prompts as architecture, not cosmetics. Stop decorating. Start designing.

Training-data curation

How to build the dataset that matters. Quality over quantity. Signal over noise.

Synthetic data generation pipelines

Making your own training data, ethically, at scale, without a cloud bill.

Eval harnesses

Measuring "better" instead of feeling it. Writing tests your model has to pass.

Prompt-to-weights reasoning

What a gradient actually does. Why your loss curve is lying to you.

Embedding-space geometry

How retrieval really works under the hood. Why your RAG gets confused.

Running your fine-tune in production

The part nobody documents. Serving, updating, rolling back when it breaks.

Gate

Prerequisites.

This one has a door. These are the things you need to walk through it.

Completed all 5 phases of AI-Trainer-MAX
A machine with more than 16GB of RAM — and preferably a GPU
Comfort at the command line. Not fluency. Comfort.
Patience. Runs will fail. That is part of the course.

If you're here without having finished MAX

Finish MAX first. OBLIVION will still be here.

Release

When does it drop?

No promised date. Shipped when it's ready, not before. MAX took roughly a year of nights and weekends. Assume similar.

TBD

Ready · Not · Scheduled

★ Star the repo to signal interest → ◉ Watch the repo for releases → @ Follow @thebardchat on GitHub →

Where this fits

The arc of the ecosystem.

OBLIVION sits between the foundation and the public platform — it's the depth layer, built for people who already finished the foundation.

ShaneBrain // local · private └── AI-Trainer-MAX // START HERE · foundation └── AI-Trainer-OBLIVION // YOU ARE HERE · depth └── Angel Cloud // public · families └── Pulsar AI // enterprise · secure └── TheirNameBrain // legacy · generational

AI-Trainer- OBLIVION

This is the sequel.

What OBLIVION will cover.

Prerequisites.

When does it drop?

The arc of the ecosystem.

Not ready for OBLIVION yet?