DEV Community: Sonam

Build an AI Audio Translator in Python on Telnyx Inference

Sonam — Fri, 26 Jun 2026 22:14:42 +0000

A lot of AI apps are starting to mix voice, language models, and generated audio.

I built a small Python example that shows that full loop:

take an audio file
transcribe it
translate the transcript with an LLM
generate translated speech

Repo: https://github.com/team-telnyx/telnyx-code-examples/tree/main/ai-content-translator-python

What it does

The app exposes a Flask API for translating spoken content.

You send it an audio file and a target language. It returns:

the original transcript
the translated text
generated translated audio

So instead of only translating text, the example shows a practical speech-to-speech style workflow.

Why this pattern is useful

This kind of flow can be useful for apps that need multilingual voice experiences, like:

customer support tools
education apps
internal enablement content
voice agents
media localization
accessibility workflows
product tutorials in multiple languages

The important part is that each step stays understandable. Speech-to-text, translation, and text-to-speech are separate pieces, so you can debug or replace one part without rewriting the whole app.

How the example works

The app uses Telnyx APIs for the voice and AI parts of the workflow.

At a high level:

Upload source audio
Transcribe the audio
Send the transcript to an LLM for translation
Generate speech from the translated text
Return text plus audio output

That gives you a clean starting point for building your own multilingual AI workflow.

Try it

Clone the repo:

git clone https://github.com/team-telnyx/telnyx-code-examples.git cd telnyx-code-examples/ai-content-translator-python

Install dependencies and set up your environment:

pip install -r requirements.txt cp .env.example .env python app.py

Then call the translation endpoint with an audio file and target language. Check the README for the exact request shape:
https://github.com/team-telnyx/telnyx-code-examples/tree/main/ai-content-translator-python

Why I like this example

It is a useful pattern for anyone building AI apps where the interface is not just text. Text-only LLM demos are helpful, but a lot of real user experiences involve audio: people speaking, systems responding, and content moving across languages.

This example keeps the workflow small enough to understand, while still showing how speech-to-text, LLM translation, and text-to-speech can fit together in one app.
The Telnyx code examples repo is also structured to be agent-readable, so coding agents can inspect the examples, understand the API patterns, and help you extend them into fuller applications.

Resources:
Code example
Telnyx Developer Docs

Build a Simple RAG App with Telnyx AI Inference

Sonam — Fri, 26 Jun 2026 22:11:23 +0000

RAG is one of those patterns that sounds more complicated than it has to be.

At its core, retrieval-augmented generation is just:

Store some documents
Embed the user’s question
Find the most relevant docs
Send those docs to the model as context
Return an answer with sources

I built a small Python example that shows that flow end to end with Telnyx AI Inference.

Repo: https://github.com/team-telnyx/telnyx-code-examples/tree/main/build-rag-with-telnyx-inference-python

What it does

The app exposes a Flask API for asking questions against a tiny in-memory knowledge base.

You send a question like:

{ "question": "Production signup broke after rotating an API key. Logs show 401 errors. What should we check first?" }

The app

creates an embedding for the question
compares it against embeddings for the sample documents
retrieves the most relevant sources
sends those sources to a chat model
returns a grounded answer plus source titles

Why this pattern is useful

A normal LLM call only knows what is in the prompt and the model’s training data. RAG lets your app answer with your own docs, policies, product information, support notes, or internal knowledge base. That makes it useful for things like:

support assistants
internal docs search
onboarding copilots
product Q&A
troubleshooting workflows
agent tools that need source-grounded answers

How the example works

The example keeps the moving parts intentionally small.
There is an in-memory DOCUMENTS list. On the first request, the app creates embeddings for those documents and caches them. When a user asks a question, the app embeds the question, compares it to the document embeddings, and sends the best matches to the model.
The answer response includes source titles, so you can see what context the app used instead of treating the model like a black box.

Try it

Clone the repo:
git clone https://github.com/team-telnyx/telnyx-code-examples.git cd telnyx-code-examples/build-rag-with-telnyx-inference-python

Install dependencies and run the app:
pip install -r requirements.txt cp .env.example .env python app.py

Ask a question:
curl -X POST http://localhost:5000/rag/ask \ -H "Content-Type: application/json" \ -d '{ "question": "Production signup broke after rotating an API key. Logs show 401 errors. What should we check first?" }'

Why I like this example

It is deliberately small, but it gives you the core pieces of a real RAG workflow:

embeddings
retrieval
source grounding
chat completion
a clean API surface

From there, you could swap the in-memory docs for a vector database, pull content from product docs, or turn it into a support assistant.
The Telnyx code examples repo is also structured to be agent-readable, so coding agents can inspect these examples and help you extend them into fuller applications.

Resources

Code example
Telnyx AI repo with skills/toolkits
Telnyx AI Inference docs

Extract Structured JSON from Messy Text with Telnyx AI Inference

Sonam — Fri, 26 Jun 2026 22:03:29 +0000

Messy text is everywhere: support tickets, lead forms, emails, contracts, incident reports, call notes, Slack messages. The annoying part is that the useful data is usually in there somewhere, but not in a shape your app can trust.

I built a small Python example that uses Telnyx AI Inference to turn unstructured text into structured JSON.

Repo: https://github.com/team-telnyx/telnyx-code-examples/tree/main/extract-structured-json-with-ai-python

What it does

The app exposes a simple Flask endpoint where you send messy text plus the fields you want back.

For example, you can send something like a customer support note and ask for:

customer name
issue type
urgency
product
next action

The model returns a structured JSON object that your app can validate, store, route, or pass into another workflow.

Why this pattern is useful

A lot of AI demos stop at “summarize this text.” That is useful, but many real apps need something stricter:

route a ticket
classify an incident
extract lead details
normalize intake forms
prepare data for a CRM
trigger automations based on extracted fields

Structured JSON makes the LLM output easier to use in actual software.

How it works

The example uses Telnyx AI Inference through an OpenAI-compatible client pattern.

At a high level:

Define the schema you want back
Send messy text to the model
Ask the model to return JSON
Validate the result before using it

That last part matters. Even when you ask an LLM for JSON, your app should still treat model output like external input and validate it.

Try it

Clone the repo:

git clone https://github.com/team-telnyx/telnyx-code-examples.git cd telnyx-code-examples/extract-structured-json-with-ai-python

Install dependencies and run the app:

pip install -r requirements.txt cp .env.example .env python app.py
Then call the endpoint with your own text.

Why I like this example

It is small enough to understand quickly, but it maps to a very real app pattern: taking messy human language and turning it into data your system can actually act on.
Also worth noting: the Telnyx code examples repo is structured to be agent-readable, so coding agents can inspect the examples, understand the API patterns, and help you extend them into fuller apps.

Resources

Code example
Telnyx AI repo with skills/toolkits
Telnyx AI Inference

Feedback welcome, especially from folks building AI apps that need reliable structured output from messy user input.