AI & automation at GreenTomatoMedia

A model response is one part of a feature. The rest includes the information it receives, the controls an operator needs, what happens when the request changes, and where a person takes over.

I joined GreenTomatoMedia as a Fullstack Developer in 2024 and now lead AI and automation development. The examples below show my hands-on work across existing products and internal tools.

I work across application logic, interfaces, background jobs, and validation. The systems shown here use Python and Django with MySQL, Redis, and Celery, alongside OpenAI and Anthropic integrations where relevant.

I built support tooling that asks AI to select an approved localized reply template. The server renders the selected template, and a person reviews it and presses Send. When no suitable template is available, the workflow records a human-review handoff.

The surrounding system includes audited settings, cost controls, background processing, and checks for changed tickets or templates. Those decisions make the review state and the next action visible to the operator.

I improved a conversational AI backend so replying to an image supplies the actual image as vision context, while replying to a voice note supplies its transcript. The quoted context remains available when the new reply also contains media.

The implementation preserves provider fallback, avoids repeatedly attaching old images on every turn, and handles unsupported media formats without failing the whole reply.

I built editable generation and translation prompts and language-specific rules into an internal content CMS. Content-aware generation and single-item translation use these controls within the editorial workflow.

Editors can adjust supported prompt behavior through the application. I also extended the AI assistance to additional content formats with their own generation and translation rules.

These contributions put AI inside a defined workflow: select an approved reply, carry the right media context, or generate content using editable rules. My responsibility is to make the surrounding software understandable and operable as well as technically sound.