Work

AI engineering, shipped to real users.

A look at what our team has built and shipped, from production AI products to computer vision and audio machine learning. This is the track record behind Sustyne's AI engineering practice.

AI Intelligence

Heartbeat

An intelligence platform that organizes people, organizations, and the relationships between them into a single searchable system. Our engineer built its features end to end across a Django and React stack and a FastAPI payments service on GCP, from Stripe subscriptions and workspace provisioning to production reliability and data-quality fixes.

  • Full-stack SaaS across Django/PostgreSQL, React/TypeScript, and FastAPI on GCP
  • Stripe subscriptions, workspace provisioning, and Google Places enrichment
  • Reliability and data-quality fixes: search self-healing, IAM signup bug, standardized geos data
AI Wellness

Junoon

An AI wellness app that pairs members with a conversational coach, habit planning, and personalized nutrition. Our engineer builds its LLM-backed coaching engine and personalization systems across a Next.js backend and SwiftUI iOS, and cut video storage and delivery roughly 5x (37 GB to 7 GB at VMAF 95).

  • Conversational AI coach (recommendation and returning-user flows)
  • Schedule and habit features, iOS and backend

Visit junoonwellness.com →

Computer Vision Founder

Every Touch

A soccer video-analytics venture founded by our AI engineer. An automated pipeline turns 90-minute matches into player-scoped highlight reels using YOLOv8 detection and ByteTrack tracking across 1.8M+ detections, with a multi-signal identification system (OSNet re-ID, jersey OCR, spatial context) that raised same-kit precision from 27% to 75%.

  • Computer vision: detection and multi-object tracking
  • Held-out evaluation and accuracy benchmarking

Visit the Every Touch site →

Audio ML Capital One

Voice Deepfake Detection

A Capital One-sponsored capstone detecting AI-generated and spoofed voice for transaction authentication. Anti-spoofing models on the ASVspoof benchmark, with an AWS architecture and model explainability analysis.

  • Wav2Vec2 and AASIST score-fusion ensemble reaching 14.87% EER on ASVspoof
  • AWS architecture and SHAP / LIME explainability
Let's talk

Your product could be the next one here.

Tell us what you're building. We'll bring the engineering and the AI judgment to match.