Skip to content
All case studies

Ironbook AI · Aug 2025 – Present

Voice AI: speech in, speech out

Engineered end-to-end voice capabilities, Speech-to-Text and voice-over, and managed the AI agent lifecycle on AWS SageMaker and Bedrock.

AIFull-stack

01Context

Ironbook AI builds AI-powered products. As a full-stack developer and AIML engineer, I work across the whole path, from models to the interface users touch.

02Challenge

Adding voice to an AI product means much more than calling a model. Speech has to become text, the AI has to respond, the response has to become natural speech again, and all of it has to be integrated, deployed and kept running.

03Approach

  1. 01

    Speech-to-Text

    Integrated and deployed STT so spoken input becomes text the AI services can work with.

  2. 02

    Agent lifecycle

    Managed AI agent lifecycles with AWS SageMaker and AWS Bedrock to build and deploy the models behind the experience.

  3. 03

    Voice-over

    Delivered voice-over capabilities so responses can be played back as speech.

  4. 04

    End to end

    Owned the full-stack integration and deployment, so the feature works as one product rather than disconnected parts.

04Architecture

  1. 1Input

    • User speech

  2. 2Understand

    • Speech-to-Text

  3. 3Reason

    • AI agents
    • SageMaker · Bedrock

  4. 4Respond

    • Voice-over
    • App UI

Simplified view of how the pieces connect.

05Stack

Speech-to-TextVoice-overAWS SageMakerAWS Bedrock

06Outcomes

  • STT and voice-over shipped end to end, from integration to deployment
  • AI agent lifecycle managed on SageMaker and Bedrock

Details are generalised to respect employer and client confidentiality. Numbers shown are the ones I can share.

Building something similar?

Tell me about it. I'm happy to talk architecture, trade-offs and timelines.

Start a conversation