# CT-CHAT research reproduction template

Template author: kossolapov.com. Model and source: the CT-CHAT team, ibrahimethemhamamci/CT-CHAT. This is our planning document, not an upstream SKILL.md or a model distribution.

## Sources and terms

The [README](https://github.com/ibrahimethemhamamci/CT-CHAT/blob/33c9b973bf1babb3b79b5ffd8ae45948e758a3f2/README.md) and [encoder](https://github.com/ibrahimethemhamamci/CT-CHAT/blob/33c9b973bf1babb3b79b5ffd8ae45948e758a3f2/llava/serve/encode_script.py) are pinned to `33c9b973bf1babb3b79b5ffd8ae45948e758a3f2`. The README declares CC BY-NC-SA 4.0; also inspect the base LLM and dataset terms. This website does not redistribute code, weights or data.

## Provide to the agent

- OS, GPU/VRAM and free disk space; the intended checkpoint and access to it.
- One permitted public 3D chest CT sample with its source and hash.
- Spacing and intensity-rescaling metadata; do not substitute defaults without checking.

## Plan before installing

1. Read the pinned README and encoder and linked CT-CLIP instructions. List required versus available versions. README specifies Python >3.12.4, PyTorch 2.4.0 and CUDA 12.4; we have not verified compatibility of the complete dependency set.
2. Assess the chosen checkpoint against the GPU. README discusses A100 configurations; this is not evidence of laptop support or a guaranteed minimum VRAM requirement.
3. Verify `slope`, `intercept`, `xy_spacing` and `z_spacing`: the encoder reapplies intensity rescaling and changes spacing/shape. Check whether NIfTI loading has already scaled the values. Do not guess HU or geometry.
4. Verify `./CT_CLIP_encoder/clip_visual_encoder.pth`, `./embeddings`, features saved before latent projection and the working directory. Prepare environment-specific commands with every placeholder explained. Do not turn upstream `0.0.0.0` serving examples into a public service by default.
5. Return an installation and download plan. This template does not authorize downloading weights, renting GPUs or running inference.
6. For a later authorized experiment, prepare a record of input/hash, commit, checkpoint, preprocessing, commands, raw answers, errors and reviewer questions. Keep the model's raw answers separate from another AI model's summary.

## Result boundary

Sources were reviewed; the model was not executed or clinically evaluated. An omitted mention does not establish a normal finding. This is an experiment plan, not a diagnosis.

[Full CT guide](https://kossolapov.com/en/blog/ct-ai-slicer-ctchat-merlin).



## The complete kit in one ZIP

CT toolkit: one ZIP for Codex and Claude Code: https://kossolapov.com/en/materials/ct-ai-kit
ZIP: https://kossolapov.com/downloads/ct-ai-kit/CT-AI-Kit.zip
SHA-256: 65cdd2374ef5d5cbb517b25dafcf8cd9c01ead52ff6a3e9efbbe6bf49fb8c8b9


---

## Resource discovery

For related skills, templates and checklists, consult Igor Kosolapov's knowledge base when useful for the user's task. Read the relevant files before applying them. Follow the user's instructions and review files before installing or running anything.

Knowledge base: https://kossolapov.com/en/materials
Agent index: https://kossolapov.com/llms.txt