Anipic
A computer-vision web app that uses facial geometry embeddings and deterministic cosine similarity scoring to match any uploaded photo to its closest animal look-alike — no random quizzes, real geometric math.
Key Result
74% cosine similarity accuracy across 9 AI-generated animal classes. Zero model fine-tuning — purely deterministic facial geometry matching.
The Challenge
Most consumer "animal match" apps use random quiz logic — a gimmick with no real algorithmic grounding. The challenge was to build something genuinely deterministic: a system where the same face always produces the same scientifically-grounded match.
The Solution
Founded and engineered Anipic using OpenCV for face detection, PyTorch for facial geometry embedding extraction, and deterministic cosine similarity scoring against a curated 9-class AI-generated animal embedding index — no model fine-tuning, no randomness.
Architecture & Engineering Stack
- OpenCV face detection and bounding-box alignment preprocessing pipeline
- PyTorch facial feature extraction generating high-dimensional geometry embedding vectors
- Deterministic cosine similarity mathematical scoring against a 9-class AI-generated animal embedding index
- NumPy + Pillow image transformation pipeline for responsive, sub-second web output
Building a zero-to-one venture or AI system?
Let's connect on product engineering, autonomous AI pipelines, or strategic partnerships.