Microleap – AI Food Detection & Nutrition Analysis

Role: Backend Developer (Freelance) • Microleap

Real-time meal recognition and nutritional breakdown platform powered by Gemini 2.5 Pro on Google Vertex AI, semantic vector search with Qdrant, and serverless Azure Functions.

// Timeline:

06/2025 — 12/2025

// Team context:

Collaborated closely with machine learning engineers, DevOps, and mobile application developers.

MODULES (SCROLL TO EXPLORE):

[CARD 01 / 07] // PROBLEM ARCHITECTURE

psychology SYS.SYS.04.01

Context & Problem Statement

Users required instant nutritional analysis (calories, macronutrients) directly from meal photographs tailored to their dietary goals. The backend had to handle unpredictable mobile traffic spikes elastically while resolving colloquial and regional dish naming variations via semantic similarity.

// ENGINEERING FOCUS:

Generative AI & HealthTech

// DEPLOYMENT SCOPE:

06/2025 — 12/2025

[CARD 02 / 07] // CORE CONTRIBUTIONS

terminal SYS.SYS.04.02

Core Engineering Contributions

›_ Architected serverless backend microservices using Azure Functions with TypeScript.
›_ Integrated multimodal vision inference via Gemini 2.5 Pro on Google Vertex AI for granular food ingredient detection.
›_ Implemented semantic similarity search using Qdrant Vector DB to normalize dish names and resolve abbreviations against nutritional databases.
›_ Enforced edge security with Azure API Management (APIM) rate limiting and Azure Entra ID (Single Sign-On).

[CARD 03 / 07] // MEASURABLE IMPACT

trending_up SYS.SYS.04.03

Measurable Impact

✓ Delivered complete image-to-nutrition analysis payloads in under 2 seconds.
✓ Scaled elastically to thousands of concurrent requests with zero idle server overhead.
✓ Automated client SDK generation and contract verification directly from OpenAPI definitions.

[CARD 04 / 07] // SYSTEM DESIGN

account_tree SYS.SYS.04.04

Architecture Highlights & System Design

› Pure serverless compute with Azure Functions reduces idle operational costs to zero.
› Qdrant vector similarity matching resolves regional dish naming variations and synonyms reliably.
› Tiered edge routing and throttling enforced through Azure API Management (APIM).

[CARD 05 / 07] // TECHNICAL SCOPE

checklist SYS.SYS.04.05

Responsibilities & Technical Scope

› Designed API contracts conforming to OpenAPI standards and authored serverless execution flows.
› Integrated AI inference pipelines with Google Vertex AI and Qdrant Vector DB.
› Codified cloud infrastructure using Terraform and automated deployments via GitHub Actions.

[CARD 06 / 07] // OUTCOMES & RETRO

verified SYS.SYS.04.06

Delivery Outcomes & Production Impact & Lessons Learned & Engineering Takeaways

verified // Delivery Outcomes & Production Impact

✓ Backend APIs supported smooth mobile beta launches without scalability bottlenecks.
✓ Entire cloud infrastructure can be provisioned reproducibly with a single terraform apply command.

lightbulb // Lessons Learned & Engineering Takeaways

› AI product speed depends heavily on asynchronous pipeline architecture surrounding the model rather than model inference latency alone.
› Contract-first development with OpenAPI specifications streamlines mobile and backend integration dramatically.

[CARD 07 / 07] // TECH SPEC & VERIFICATION

layers SYS.SYS.04.07

Technology Stack

Azure FunctionsTypeScriptGoogle Vertex AI (Gemini 2.5 Pro)Qdrant Vector DBAzure APIMAzure Entra IDTerraformGitHub ActionsOpenAPI

[DOSSIER COMPLETE • SYS.04] ● 100% PRODUCTION VERIFIED

arrow_back ← Back to all projects [SYS.04] • Microleap – AI Food Detection & Nutrition Analysis