Health & wellbeing

Vitamin AI

From scattered documents to personal context.

Lab-document understanding, nutrition tracking and contextual information retrieval in a mobile-first product.

FlutterTypeScriptNestJSMySQLPostgreSQLRedis

MY ROLE

Product & Full-Stack Engineering

BUILT FOR

People organizing nutrition information and lab documents

01 / THE CONTEXT

From scattered documents to personal context.

Lab documents, daily nutrition records and reference material often live in separate places. Building context requires both structured records and a way to retrieve relevant supporting information.

What I built

I built a Flutter experience with a NestJS backend, document extraction jobs and a retrieval layer. Application records and vector retrieval have distinct storage responsibilities.

02 / PRODUCT CAPABILITIES

Lab-document processing

Background extraction workflows turn uploaded documents into information that can be organized within the application.

Nutrition & check-ins

Mobile tracking and check-in workflows help users maintain personal context over time.

Contextual retrieval

A PostgreSQL / pgvector retrieval layer complements MySQL application storage for reference-oriented AI workflows.

Under the surface.

Follow a real product workflow through its application boundaries. Select any step to explore the implementation.

ENGINEERING IN MOTION

A lab document becomes a structured record.

The asynchronous extraction path handles PDFs and images, reuses content-based results and reports completion back to the app.

STEP 01 / USER & FILE CONTEXT

Upload record

A lab upload identifies the user, storage object and file type. A queued job carries those identifiers into background processing.

WHY IT MATTERS

Keep document extraction asynchronous, distinguish cached and fresh analysis, and make both success and failure visible to the mobile client.

Conceptual flow based on the implementation. Animated connections illustrate data movement, not live traffic.

04 / ENGINEERING JUDGMENT

Built around the constraints.

Separate application data from retrieval data. Keep lengthy extraction work asynchronous, and describe AI-assisted information as support rather than a medical diagnosis.

Discuss the engineering ↗