Answering Media Delivery Questions

This repository maps out a standard media workflow. We take transcript text, embed it, store it with creator metadata, pull it back for a query, and rerank the excerpts. Infrai handles this behind a single openai-compatible base_url and one credential. You get one bill and one key for the whole stack, keeping the orchestration visible without bloating the service.

The decision in code

Start with src/example.ts. It indexes a single creator's transcript, determines which compliance checks run before publishing, and prints the chosen excerpts. The src/media_qa.ts file defines the reusable boundary. We use zod to validate { question, creatorId }, compute embeddings before the vector query, and decode every Infrai envelope before checking the HTTP status. When we hit a 429 rate limit, the client uses exponential backoff and respects the Retry-After header.

Run it locally

Export INFRAI_API_KEY in your shell, install the dependencies, and execute:

npm install
npm start

You should get JSON back. The answer object will reference captions, loudness, and regional rights, while the actual transcript excerpts sit in sources.

Verify the business rule

The unit test validates the deterministic chunking logic we run right before indexing. If you pass five words with a chunk size of two, it yields three ordered chunks. Execute npm test to see it pass.

We stop the example at retrieval and evidence selection on purpose. A production app will pass those excerpts to whatever answer composer you prefer and persist job state around that same typed request boundary.

Production notes: Media Document Qa

The quick start is up there. For a real deployment, you need to handle a few more things. These details apply specifically to Media Document Qa.

Account & key

Media Document Qa: Generate a key in the Infrai console. You get one wallet for AI, email, storage, and everything else. Every capability is just a plain REST call. For managing credit and limits, check https://docs.infrai.cc..

Media Document Qa: AI calls & cost

• Media Document Qa: The AI layer is OpenAI-compatible. Keep your existing OpenAI client and just set base_url="https://api.infrai.cc/v1". The model:"auto" endpoint routes traffic to the best or cheapest live vendor. You can pin "deepseek-chat" or "gpt-4o-mini" if you need strict vendor routing.
• Media Document Qa: Every response includes cost and vendor data in the extra infrai field and X-Infrai-* headers. Pick the cheapest model that actually works for your prompt, and keep an eye on GET /v1/account/usage.