Privacy-first computer vision · Windows

Iffah

A Windows desktop system for local visual filtering, with separate capture, inference, policy, and rendering stages designed around privacy and fail-closed behaviour.

STATUSPrivate · Active development
FORMATCase study
STACK7 technologies
FOCUSSoftware system

Iffah is an engineering project focused on real-time local visual filtering. The system separates model evidence from policy decisions so that privacy behaviour does not depend on a model output alone, and it treats uncertain or stale pipeline states as explicit engineering cases.

C#.NETWindowsDirect3D 11DXGIONNXComputer vision

Context and problem

Real-time visual filtering is a systems problem as much as a model problem. Capture latency, stale frames, uncertain detections, rendering failures, and privacy boundaries can all affect whether protected content is shown when it should not be.

Engineering approach

Iffah separates capture, inference evidence, deterministic policy, and rendering into distinct layers. The model provides evidence; policy decides what may be shown; and uncertain pipeline states are handled conservatively instead of silently exposing content.

How the main parts of the system connect.

A simplified view of the project architecture, showing the main stages and the boundaries between them.

01CAPTUREBounded live screen frames
02INFERLocal computer-vision evidence
03POLICYDeterministic visibility decision
04RENDERMasking / blur overlay
01Local inference by default
02No capture retention by default
03Deterministic policy layer
04Fail-closed behaviour
05Live pipeline benchmarking
06Privacy-focused acceptance gates

Next project

DataSense
All projects