Gates Foundation

gatesfoundation

A workspace for the Gates Foundation DPI, FairFoward and Gooey teams focused on evals for low-resource languages plus the home of our Agriculture advisory work e.g. https://gooey.ai/ageval

352 Public Workflows
8 Members

This workflow processes Kikuyu audio inputs by first transcribing and translating them into English. After translation, an advanced language model (LLM) generates detailed responses or answers based on the translated English text. The workflow runs two different model pipelines for comparison, evaluates their outputs with dedicated evaluation workflows, and saves the results for further analysis. This setup helps determine which pipeline provides the most accurate and effective end-to-end solution for Kikuyu audio understanding and intelligent response generation.

🦾

10mo ago

Public

This workflow is designed specifically to measure end-to-end latency for voice-based AI interactions, rather than to provide full answers. Incoming audio (in Kikuyu or other selected languages) is transcribed, processed by an AI assistant with a maximum output of 10 tokens, and then synthesized back to audio. This setup allows for reliable benchmarking of system response time across the transcription, AI, and text-to-speech components.

10mo ago

Public

This workflow is designed specifically to measure end-to-end latency for voice-based AI interactions, rather than to provide full answers. Incoming audio (in Kikuyu or other selected languages) is transcribed, processed by an AI assistant with a maximum output of 10 tokens, and then synthesized back to audio. This setup allows for reliable benchmarking of system response time across the transcription, AI, and text-to-speech components.

10mo ago

Public

AI assistant for African farmers that accepts Kikuyu audio, transcribes and responds in English. Powered by GPT-4o Audio for accurate speech recognition and practical farming advice.

10mo ago

Public

This workflow converts spoken Kikuyu audio into English and Kikuyu text responses, powered by state-of-the-art speech recognition, translation, and GPT-4.1. Designed to assist African farmers, it provides accurate, up-to-date information in both languages. Use it to transcribe, translate, and interact with agricultural knowledge hands-free.

Improved the workflow description for clarity and detail.

10mo ago

285 runs

Public

Here we take an input sample of Kikuyu audio questions (their expert transcription + English translations) AND their expert created answers in Kikuyu and English. We then compare chained models (ASR + LLM + TTS) vs single models (GPT4o Realtime) to determine which are faster and semantically closer to the golden answers.

🦾

10mo ago

Public

Here we take an input sample of Kikuyu phrases (and their expert transcription and English translations) and compare the top Speech Reco and Translation models. As of Mar 25, 2025, we've found that Meta's MMS-Large works best for Kikuyu with Ghana NLP's Kikuyu to English translation services.

. Which AI model actually works best for your needs?
Upload your own data and evaluate any Gooey.AI workflow, LLM or AI model against any other.
Great for large data sets, AI model evaluation, task automation, parallel processing and automated testing.
To get started, paste in a Gooey.AI workflow, upload a CSV of your test data (with header names!), check the mapping of headers to workflow inputs and tap Submit.
More tips in the Details below.

Compare Kinyarwanda Speech Recognition and Translation to English with this run.

Compare Swahili Speech Recognition and Translation to English with this run. We have 10 variations in this. Including the latest Seamless M4T v2, MMS, Azure, Whisper v3 and more.