Assessment Pipeline: Add audio/video support

[kartpop]

Is your feature request related to a problem?
The AI Assessments pipeline currently lacks support for audio and video modalities. This limits the assessment's effectiveness and prevents a comprehensive analysis of inputs.

Describe the solution you'd like

• Extend multimodal support to include audio/video in AI Assessments.
• Experiment with Gemini for video handling and explore methods with OpenAI/Anthropic.
• Sample an audio/video dataset from partners, run the pipeline, conduct human evaluations, and iterate.
• Support a language mix of ~50–60% English, ~10–15% Tamil, and a strong presence of South Indian languages like Telugu.

Original issue

Context

Extend multimodal support to include audio/video in AI Assessments.

Investigation

• Video: Gemini does appear to support video directly. Experiments to be done to figure out reasonable way of handling video in Open AI / Anthropic (sampling frames + full audio transcript ?)

Approach / acceptance criteria

• Sample audio/video dataset from partners, run pipeline, run human evals, iterate. Building the tech is easy; evaluation is the bottleneck.
• Language mix to support: ~50–60%+ English, Tamil ~10–15%, strong South Indian language presence, Telugu