PERSONAL PROJECT / ENGINEERING CASE STUDY

FabricAssist.

Building safer AI-assisted garment care.

FabricAssist is a cross-platform project for garment owners and professional cleaning services. It collects garment photos, extracts source-backed evidence with AI, and validates actionable guidance through deterministic Python rules.

Current project status

Active development / prerelease. Live provider and production activation remain gated by testing, privacy, licensing and domain-expert review.

01 / THE ENGINEERING CHALLENGE

The engineering challenge

How do you make vision-driven suggestions useful while preserving uncertainty, source traceability and predictable guardrails?

02 / SYSTEM DESIGN

System architecture

A shared Expo client talks to a FastAPI backend. Durable state lives in PostgreSQL, background work runs through workers, and images live in private object storage. Vision-model adapters are replaceable.

FABRICASSIST / ARCHITECTURECLIENT → API → WORKER → DATA
SYSTEM EXPLORERSELECT A SERVICE ↗

Photos → API → Analysis → Validated guidance

02
FastAPI

Typed API boundaries authenticate requests and persist analysis intent before dispatching background work.

Architecture diagram is illustrative and intentionally simplified.

03 / ENGINEERING DECISIONS

Selected technical decisions

01

Structured evidence first

Source-referenced image evidence and typed schemas precede any actionable recommendation.

02

Deterministic safety gate

Versioned Python rules validate or reject AI-generated care claims.

03

Asynchronous processing

Durable job state, workers, retry behavior and provider concurrency management.

04

Portable infrastructure

Adapters isolate AI models, job dispatch, caching and storage from application logic.

05

Operational readiness

Automated tests, OpenAPI contracts, logs, telemetry and release procedures.

04 / TECH STACK

Technology

PythonFastAPIPydanticPostgreSQLCeleryValkeyReact NativeExpoTypeScriptDockerMinIO / S3OpenTelemetryAI Vision APIs

No production or customer-outcome claims are made for this prerelease project.

Back to portfolio