An archive that reads, translates and broadcasts itself.
Thirty years of design thinking: 450 talks and 612 hours of transcribed conversation, kept on the record, made legible in any language, and carried across Africa and the open web. This is where Design Indaba is headed: not a bigger website, but a living archive that does more of the work itself.
Designed to run on AWS · Bedrock · Transcribe · Translate · Elemental
- 450
- Talks
- 612
- Hours transcribed
- 6
- Languages
- 84
- Countries reached
What readers experience.
Public-facing intelligence: shown in the prototype today, with more on the roadmap.
In the prototype
Ask anything, in any language
Natural-language semantic search across 612 hours of transcribed talk. Ask a question the way you'd ask a person.
In the prototype
Explore by question
Query a single talk's content and jump straight to the moment in the video where it's answered.
In the prototype
Read or watch in any language
Transcripts and subtitles in English, Português, Kiswahili and Français, with isiZulu and isiXhosa in the queue. Africa-wide reach, by design.
In the prototype
Translated search
Search in Portuguese, find an English talk. The archive speaks more languages than any single visitor.
In the prototype
Semantic related content
Connections drawn from transcript themes, not tags. They surface the non-obvious talk three years and two disciplines away.
In the prototype
Short-form clips
Every long talk yields vertical clips for YouTube Shorts, Instagram and TikTok: a funnel back to the full archive.
Proposed
Curated journeys
Say what you want to learn, and the archive sequences a path through the talks: a course, with the connective notes written for you.
Proposed
Follow a thread
Pick an idea (repair, soft power, biomimicry) and watch how the thinking on it evolved across thirty years of the festival.
Proposed
Watch it spoken
Not just subtitles: AI-voiced dubbing of a talk into Kiswahili, isiZulu or Português, so you can listen in your own language.
Proposed
The data-light read
On a thin connection? A rich, illustrated read of any talk, generated from the transcript, so the ideas arrive without the bandwidth.
What editorial experiences.
The CMS side: the part that makes this a platform, not a website. Shown as annotated mockups; built for real in the next phase.
Sanity Studio · Curator AI panel
Curator AI
An editor narrates raw context (“this was filmed the week of the water crisis; the speaker had just been released from custody”), and the AI returns a polished curator's note in the editor's own voice, ready to publish.
Sanity Studio · built in phase 2
Sanity Studio · suggestion sidebar
AI tagging & linking
While editing a talk, the sidebar proposes related talks, topics to tag, and speakers named in the transcript, each one auto-linkable with a click.
Sanity Studio · built in phase 2
Sanity Studio · translation queue
Translation queue
Every document's translation status at a glance (EN, PT, SW, FR), with the ability to prioritise, regenerate or hand-edit any language.
Sanity Studio · built in phase 2
Sanity Studio · clip review
Clip review
The AI proposes candidate short-form clips with reasoning (“key quote”, “audience reaction”, “visually striking”), and the editor approves which ones go live.
Sanity Studio · built in phase 2
Sanity Studio · distribution
Distribution dashboard
One editorial publish, fanned out to YouTube, Spotify and Instagram. Performance flows back into one view.
Sanity Studio · built in phase 2
Publish once. Everywhere.
One editorial action fans out across the open web. Each platform's format is generated automatically, every version linking back to designindaba.com.
Editorial publishes once
YouTube Shorts
clipped to 60s
links back to full talk
Spotify
audio + spoken intro
podcast feed
square, auto-captioned
links in bio
How AWS could power the whole archive.
What it takes to make thirty years read, translate and broadcast itself, and the AWS service that would power each piece. Tagged Need now, would adopt, or exploring.
Store & serve the archive
A durable home for thirty years of video masters, images and transcripts: the vault itself.
A global CDN so a talk loads as fast in Lagos or Bogotá as in Cape Town. Reach is the whole point.
Transcode each talk once into adaptive streaming and every channel's format. Publish once, deliver everywhere.
Just-in-time packaging, and live-streaming a future festival straight into the archive.
Read every talk: the editorial intelligence
Turns 600+ hours of talk into timestamped transcripts: the foundation every other feature stands on.
Machine translation into Português, Kiswahili and Français, with isiZulu and isiXhosa next. Africa-wide reach, by design.
Managed foundation models for the curator's voice: the short version, ask-this-talk, the semantic threads. No model ops to run.
Entity, topic and key-phrase extraction to auto-tag talks and build the subject index.
Scene and segment detection to find the visually striking moments and auto-cut short-form clips.
Natural text-to-speech for audio editions, accessibility, and spoken podcast intros.
Search & discover
A vector store for Bedrock embeddings. Ask a question in any language, find the talk that answers it.
Managed retrieval so ask-this-talk answers stay grounded in the transcript, with citations.
A managed intelligent-search layer if we want relevance tuning out of the box.
Broadcast: publish once, everywhere
The pipeline: one editorial publish fans out into serverless steps that transcribe, translate, embed, clip, package and distribute.
Each publish emits an event, so every downstream transform triggers itself.
Queues the heavy jobs so a whole festival's uploads process smoothly, not all at once.
Run & watch the platform
Host the Next.js front end on managed edge infrastructure, beside everything else.
Metadata, translation status and distribution analytics on a pay-for-use data layer.
Observability for the pipeline, and the guardrails that keep inference cost capped.
Why back it
- A flagship Global-South culture archive built on Bedrock and Elemental: a reference customer for AI-for-good in a market AWS is investing in.
- Thirty years of design thinking, made reachable in every African language: the kind of story that headlines a keynote.
- Inference scales with the editorial calendar, not traffic: a clean, sponsorable cost profile where credits go far.
Responsible scale
The AI features run with smart caps, caching and pre-computation: semantic embeddings and translations are generated once at publish time, not per request. Inference cost scales with the editorial calendar, not with traffic. Responsible scale, not runaway bills.
The partnership
A cultural archive, preserved and broadcast — at the speed of the open web.
The reader experience is demonstrated in this prototype. The editorial intelligence and continent-wide distribution come to life with an infrastructure partner like AWS. That is the conversation this page is here to start.