I am building the first working prototype of ZENVIA’s AI-powered smart mirror and need a computer-vision specialist to own the core software stack for the next 3–4 weeks. The mirror will run on Windows and must: • Log users in via face recognition with high accuracy—false accepts/rejects should be rare enough that manual override is almost never needed. • Overlay chosen outfit images in real time so the user can “try on” clothes virtually (initially a simple 2-D overlay aligned to the torso; we will refine later). Tech stack Python 3.x with OpenCV is non-negotiable. If you can fold in MediaPipe for landmark detection that will be a welcome bonus and will likely boost performance. Data I do not yet have any datasets—neither for face recognition nor for outfit imagery—so part of the assignment is advising on, sourcing, or quickly curating open-source data that will let us reach production-quality results without violating licenses. Scope of work 1. Design and train (or fine-tune) a high-accuracy face recognition model suited to a single-user device. 2. Build a lightweight Windows application that detects the user’s face, verifies identity, and immediately unlocks the interface. 3. Implement a modular overlay pipeline: load PNG outfit layers, align them to the live camera feed, and render at interactive frame rates. 4. Package everything so that it can run on commodity hardware with a standard webcam. Deliverables • Source code with clear setup instructions • Pre-trained model files • Minimal GUI demo (Python/Qt or similar) • Short video demo showing login and try-on flow • One-page hand-off document describing any further training steps Timeline & milestones Week 1: Data strategy + baseline model Week 2: Face recognition integrated into Windows demo Week 3: Outfit overlay module + performance polish Final days: Testing, bug fixing, hand-off Budget is INR 60k–80k; I’m happy to split it across the milestones above. Please share links or repos from past AI/computer-vision projects—particularly face recognition or AR/try-on work—so I can see your style and code quality. Looking forward to collaborating on this rapid build.