Skip to content
heiSay hello

NutriTrack — AI workflows at FCAJ

Recipe generation, contextual coaching and structured responses for a nutrition app, built during First Cloud AI Journey.

FCAJ program · Certified OJT, Jan–May 2026

FCAJ Cloud Intern · Team NeuraX · AI workflows

Source code & documentation
NutriTrack solution architecture from the internship workshop, showing the wider team system.
NutriTrack solution architecture from the internship workshop, showing the wider team system.
Certified OJT period
Jan–May 2026
AI workflow
Context → JSON
FCAJ program team
Team NeuraX

The problem

NutriTrack connects meal logging, ingredient information and nutrition features. A useful AI response needs the user's context and an output the app can actually consume, rather than an unrestricted block of generated text.

My part

I worked as a FCAJ Cloud Intern in Team NeuraX in the First Cloud AI Journey program at AWS Vietnam. My AI work included recipe-generation and coaching workflows, with DynamoDB context supplied to Amazon Bedrock through Lambda functions.

The internship worklog records prompt design, structured JSON responses, integration with the application developer, and timeout handling. I also documented the implementation for the technical workshop.

Through the workflow

  1. Read relevant user or ingredient context from DynamoDB.
  2. Pass that context to a Bedrock-backed Lambda workflow.
  3. Return structured JSON for recipe or coaching features.
  4. Connect those responses to the app through AppSync.

The architecture above comes from the internship workshop. It shows the wider team system; my contribution focuses on the AI workflows and their integration.

Period and evidence

I completed the certified OJT training period in January–May 2026.

The internship report contains weekly worklogs and workshop material. The Week 8 worklog documents recipe integration; Week 9 covers contextual coaching.

Project scope

The worklogs document implementation, integration demos and workshop delivery. Production usage, operating costs and model performance were not measured during this training project.