CCF Tech for Good Hackathon 2025

Project Overview

Built a social-impact prototype that generates audio descriptions for films to improve accessibility for visually impaired audiences. We mapped user needs, defined success metrics, and delivered a working demo that turns “silent” gaps into concise scene narration while preserving dialogues.

What I Did

• Framed the problem with stakeholder interviews and benchmarked current manual workflows for accessible films.
• Designed an end-to-end pipeline: speech/silence detection → scene understanding → narration drafting → voiceover mix.
• Implemented clean prompts/flows for scene summarisation and tone control; added basic guardrails and fallbacks.
• Built evaluation rubrics (clarity, timing, non-spoiler tone) and ran quick user-style pilots for iteration.
• Packaged a lightweight UI for one-click generation and manual tweaks to support accessibility editors.

Reflection

Hackathons reward clarity over complexity. Scoping around “insert narration only when dialogue is silent” forced disciplined engineering and better UX decisions. The key lessons for me: 1) model choice matters less than crisp prompts and timing control; 2) accessible design needs measurable guardrails (no spoilers, readable pace); 3) a small, end-to-end demo with editing hooks is more valuable than a perfect model benchmark.