Problem
Vision-assisted capture must handle uncertain readings without treating a model suggestion as confirmed data.
Project
1Pour (formerly BarLens)
Production applied AI integration using Gemini vision, Next.js, Supabase, and human-confirmed data workflows.
Approach
I connected Gemini suggestions to a Next.js workflow with human confirmation, deterministic checks, and Supabase-backed records.
Result
The production applied AI workflow keeps uncertain captures reviewable and reporting grounded in confirmed records.
What I work on
At 1Pour (formerly BarLens), I integrated Gemini vision suggestions into a Next.js capture flow. Users confirm uncertain readings before they become operational records.
Engineering principles
- Keep model-assisted decisions reviewable.
- Make uncertainty visible to users.
- Keep media private and records auditable in Supabase.
- Test the full workflow in sandboxes, not only the model output.
- Prefer simple, reliable product behavior over hidden automation.
Source availability
The source repository and customer data are private. This page describes the engineering approach without exposing customer-specific details.