Skip to main content

Core Use Cases

Kaapi is designed around three production workflows currently — more to follow.

Build a knowledge-base chatbot​

Anchor a chatbot on your documents and ship it through Glific or a custom client.

The flow:

  1. Upload documents
  2. Create a vector store (collection)
  3. Build an AI config
  4. Quick-test in console
  5. Curate a golden dataset
  6. Run evaluations
  7. Iterate on prompts and parameters
  8. Lock the config
  9. Go to production through llm/call/

Build a speech-to-speech bot in local languages​

Two patterns are supported.

Pattern A — Compose your own pipeline​

Stitch STT, text + filesearch, and TTS yourself across three llm/call/ requests. Maximum control and per-step evals.

Pattern B — Use the chain template​

Use the llm/chain/ speech-to-speech template — voice in, voice out, single endpoint, intermediate text answers returned via webhook callbacks.

Build an AI assessment pipeline​

Score multimodal submissions at scale with rubric-based prompts and structured output.

The flow:

  1. Upload a submission dataset
  2. Map columns to inputs, attachments, and ground-truth scores
  3. Build an AI config encoding the rubric
  4. Submit a batch run via assessments/
  5. Compare results across up to four configurations
  6. Iterate on rubric phrasing and model choice
  7. Re-run
Detailed guides coming soon

Walkthroughs for each use case are on the roadmap. In the meantime, reach out at kaapi@projecttech4dev.org for a guided demo.

Ready to try it? Head to Getting Started.