All selected work
02 / PRODUCTION AIIn production · Since April 2026

Useful answers start with the right information.

DEVEJI client application · Design & implementation
Tool callingVoice & textProduction
A CONVERSATION. CONNECTED.
VOICE & TEXT
APP CONTEXT
✳TRAVEL AI
TRAVEL DATA
PLACES & HOTELS
Ask. Explore. Go. IN PRODUCTION
01 / THE CHALLENGE

Bring the right context into the conversation.

The assistant needed to answer travel questions inside an existing client app, using our travel database and app knowledge. It also needed to work in both text and voice, respect access permissions and use only the user context explicitly provided by the application.

02 / THE DECISIONS
01

Connect the assistant to travel tools

Internal tools retrieve hotels, attractions and interesting places from the travel database. DeepSeek, accessed through OpenRouter, powers the assistant; ElevenLabs provides the voice experience.

02

Keep user context explicit

The app supplies the user data that the assistant may use. The agent does not independently retrieve user data. This keeps that boundary in the application’s control.

03

Design the production controls

OpenAI moderation checks messages, while usage limits and access permissions control use of the assistant. These controls were part of the production implementation alongside the conversational experience.

03 / THE RESULT

What came out of it.

The assistant is live in the client’s production app. Users can choose text or voice to ask travel-related questions, supported by database tools and the context the application explicitly provides.

Client identity and application details are kept private. The visual above is a simplified architecture illustration.

WHAT I TAKE FORWARD

“The conversation is only one part of a production assistant.”

NEXT PROJECT / 03Better tools. Better workflows.