hect.dev

Daegu, South Korea

Hector Acosta

ML Engineer, Computer Vision & Edge AI

I build computer vision models and ship them to the edge. I lead the AICU Gait Scanner, from pose and gait models to optimized on-device builds for iOS and Android, and the services behind them.

8+ years of software engineering, 3+ years of applied ML. MSc Computer Science, Kyungpook National University.

Portrait of Hector Acosta

Experience

Where I've worked

  1. Nov 2024 – Present

    AICU

    Machine Learning Researcher, Gait Scanner lead

    • Lead the AICU Gait Scanner: computer vision models for real-time gait analysis used by clinicians.
    • Own the path from training to shipping: optimized edge inference with Core ML on iOS and CUDA/C++ accelerators, native Android and iOS apps.
    • Connect the on-device stack with Python and Ruby on Rails services for regulatory-ready rollouts.
  2. Sep 2021 – May 2025

    BetterUp

    Software Developer, part-time contractor

    • Built internal integrations with Microsoft Teams, Slack and other third-party platforms (Ruby on Rails, JavaScript).
    • Helped build Microsites, the alternative payment platform that handled thousands of partner transactions.
  3. Sep 2019 – Sep 2022

    Kaleyra (formerly mGage)

    Software Engineer Specialist

    • Designed a multimedia transcoding system and high-throughput messaging applications at mGage.
    • After the Kaleyra acquisition, moved PHP monoliths to multithreaded Go services (Java, Rails, Go, PHP).
  4. Nov 2018 – Nov 2019

    Pierpont Global

    Co-founder, lead software developer

    • Co-founded an online car auction platform and led a team of four building the full stack (Rails, JavaScript, Java).
    • Rebuilt in 2026 as a live demo with ML pricing and photo-assisted listings.

Research

Efficient detection for fixed cameras and edge devices

MSc thesis

Gated YOLO

A dynamic gating mechanism inside YOLOv6 with scene-specific pruning, to cut compute for real-time detection on edge devices without losing accuracy.

Code →

JICCE 2023, co-author

A scene-specific object detection system utilizing the advantages of fixed-location cameras

Edge AI detection for fixed cameras on embedded hardware.

Paper →

Dataset, 2023

ROAD-SEC dataset

13,221 images and 166,192 labeled objects from fixed security cameras over highways and roads, in varied weather and light.

Dataset →

Projects

Things I build and run

Everything here runs on my own k3s cluster: Helm charts, GPU inference with Ollama, VictoriaMetrics and Grafana alerting.

Pierpont

Live car auctions with ML pricing (LightGBM, conformal intervals), CLIP photo checks in int8 ONNX and weekly retraining behind a promotion gate.

  • Rails 8
  • Hotwire
  • LightGBM
  • ONNX
  • FastAPI
  • Helm

finance-dashboard

Self-hosted finance app with an MCP endpoint, so an AI agent can manage it through task-level tools behind a token and a NetworkPolicy.

  • Fastify
  • Next.js
  • Flutter
  • MongoDB
  • MCP
  • Helm

CRAFT Line Extractor

Document line segmentation built on CRAFT word detection: deskewing, polygon grouping into lines, and a web API.

  • PyTorch
  • OpenCV
  • Flask
  • Next.js

affine-ollama-proxy

Speaks the OpenAI Responses, Chat Completions and Anthropic Messages APIs and runs them on local Ollama models. Tested at 95% coverage, signed images with SBOMs.

  • FastAPI
  • Ollama
  • Cosign

Rock Paper Scissors AI

Hand gesture classifier trained from scratch, served as a playable game.

  • TensorFlow
  • Python

Toolbox

What I work with

ML & CV

  • PyTorch
  • TensorFlow
  • YOLO
  • Pose estimation
  • OpenCV
  • Core ML
  • CUDA/C++

Platforms

  • iOS
  • Android
  • Edge inference
  • Ollama
  • MCP

Backend

  • Python
  • Go
  • Ruby on Rails
  • TypeScript
  • Java

Infra

  • Kubernetes (k3s)
  • Helm
  • Docker
  • VictoriaMetrics
  • Grafana
  • GitHub Actions

Kyungpook National University · MSc, Computer Science · Daegu, South Korea

INTEC · Undergraduate studies · Santo Domingo, Dominican Republic

Open to ML engineering and research roles in the US.

The fastest way to reach me is email.