Alfie - research jam

Overview

Generate AI models based on user constraints
User : Wants an AI model, might not know what that entails
- Pitfalls : Bias, loads of extra time, no proper evaluation or explainability
Extra sources of information to help guide the user
Use cases - Website Accessibility checker, Compliance screening, Driver drowsiness detection
Spot the problem? :o (remember this is AutoML)

Initial idea : use an LLM to create, train and deploy models. But sustainably, cheaply and as fast as ChatGPT :)
New idea:
- Transfer learning (Hugging face, other sources) (for vision atleast)
- Tabular - AutoGluon (for now atleast)
- Basically function calling a model + data (image/prompt)
- Eg: Website accessibility - Alt text checker model (qwen2.5vl for now) + Using LLMs to check guidelines (gemma3:4b for now)

Missing ARIA attributes
The page lacks ARIA attributes to enhance accessibility for users with assistive technologies. Consider adding roles and attributes to elements like the container div to provide semantic meaning and improve navigation.
Example:
```
<div class="container" role="region" aria-label="Content Area">
```
Color Contrast
Evaluate the color contrast between text and background colors for sufficient readability, especially for users with low vision. While the current colors are generally acceptable, automated tools can identify potential issues.
Use a contrast checker to verify compliance with WCAG guidelines (at least 4.5:1 for normal text and 3:1 for large text).
Keyboard Navigation
Ensure that all interactive elements (links, buttons, form fields) are accessible and navigable using the keyboard. Check that the tab order is logical and predictable.
Link Text
Review link text to ensure it’s descriptive enough for users to understand where the link leads. Avoid generic phrases like “click here.”
Example:
```
<a href="#">Learn more about the Rubin Observatory</a>
```

Image alt text
The alt text "This is a dog" is inaccurate and misleading. The image shows a monkey, not a dog. It should accurately describe the content.
Example:
```
<img src="https://media.cnn.com/api/v1/images/stellar/prod/160107100400-monkey-selfie.jpg?q=w_1160,c_fill/f_webp" 
     alt="Monkey selfie">
```
Color Contrast
While not explicitly stated, it’s recommended to verify text-to-background contrast for readability. Use a contrast checker to ensure compliance with WCAG guidelines (minimum 4.5:1 for normal text, 3:1 for large text).

	model	score_test	score_val	eval_metric	pred_time_test	pred_time_val	fit_time	pred_time_test_marginal	pred_time_val_marginal	fit_time_marginal	stack_level	can_infer	fit_order
0	NeuralNetFastAI	0.942	0.94368	accuracy	0.0114009	0.00471711	4.5952	0.0114009	0.00471711	4.5952	1	True	3
1	LightGBM	0.941	0.952441	accuracy	0.0684869	0.023917	6.77049	0.0684869	0.023917	6.77049	1	True	5
2	WeightedEnsemble_L2	0.94	0.953692	accuracy	0.0863461	0.0385489	6.78659	0.00211716	0.000275135	0.012337	2	True	6
3	LightGBMXT	0.932	0.941176	accuracy	0.0894179	0.0278277	13.5024	0.0894179	0.0278277	13.5024	1	True	4
4	KNeighborsDist	0.2235	0.244055	accuracy	0.0157421	0.0143569	0.00376296	0.0157421	0.0143569	0.00376296	1	True	2
5	KNeighborsUnif	0.201	0.236546	accuracy	0.0157351	0.0151029	5.04704	0.0157351	0.0151029	5.04704	1	True	1