Manlio Barajas, PhD

Applied computer vision

Showcase demo

Superpixel Mesh / Aligning a mesh to any image

Profile

Computer vision engineer working on 3D reconstruction with some fullstack dev skills. Mostly working for US startups solving challenges for which computer vision has a solution.

Value proposition:

Production grade code. No need to refactor everything upon completion. Research, code and productionize as it makes sense.

Computer vision + full stack. Right time-to-market but some budget constraints? Fullstack duties covered too.

Culture and communication. 6 years of experience with US based companies. Main timezone is CST, yet flexible.

Complexity not a problem. Experience architecting and implementing year long projects smoothly. Coaching included.

Code running in production across the 5 continents

Skills

Classic Computer Vision

  • Structure from Motion
  • Bundle adjustment
  • Feature matching
  • Intrinsics and rolling shutter model
  • GPS and control point optimization
  • SDF Fusion
  • Sensor fusion
  • Pointcloud registration, ICP
  • Multiview stereo
  • Color normalization
  • Surface reconstruction
  • Texture mapping

Machine Learning

  • Frameworks Pytorch, HuggingFace transformers, drjit
  • Networks Transformers, Roomformer, SALAD, Metric3D, Oneformer, DinoV2, VGGT
  • VLM/LLM Gemma3 (fine tuning)

Rendering & Visualization

  • Frameworks mitsuba, pytorch3d
  • Methods Gaussian splatting, inverse rendering, photometric optimization

Software Engineering

  • Languages C++, TS/JS, Python, Bash, Ruby, C#
  • Toolchain CMake, GTest, pybind11, pytest
  • Web React, SvelteKit, emscripten
  • Mobile Android (classic and Compose)
  • Infra AWS, GCP, Vercel, Neon, Github Actions
  • Evaluation Pachyderm, ArgoWorkflows

Work experience

Construction and insurance platform

San Francisco, CA

Scaling a 3D reconstruction pipeline now used for millions of reconstructions. Textured meshes for interior walkthroughs, a virtual experience that replaced a third-party vendor, and a WebAssembly-based bundle adjustment tool for CAD refinement.

Takeaways: Shipping ML-heavy pipelines at scale, balancing accuracy with performance on mobile AR captures. 2 patents.

Knee Replacement company

Contract, Remote

React-based CT segmentation editor for 3D visualization and a queue backend to run ML inference for reconstructing tibia and femur models. Part of an FDA-approved surgical planning system.

Takeaways: Working with medical imaging constraints and the rigor needed for FDA-grade software.

Drone company

Guadalajara, Mexico

Sub-5cm accurate photogrammetry pipeline from scratch, shipped to edge devices on every continent. Fastest SfM solution for drone mapping in the market. Optimized for GPU on ARM (Jetson) and x86, processing billions of points in under 7Gb. Later, contractor work on multi-sensor SLAM (LiDAR, GNSS, IMU, camera).

Takeaways: Building a product from zero, leading a team, and squeezing performance out of constrained hardware. 3 patents.

Stealth startup

Guadalajara, Mexico

React/Electron surgical planning app for a robotic system that produced 3D representations from CT scans for implant placement.

Takeaways: Bridging frontend UX with heavy 3D computation in a medical robotics context.

PhD research

Monterrey, Mexico & Edmonton, Canada

Visual tracking and 3D pose estimation for UAV navigation. 4 published research articles including one at IROS. Research stay at University of Alberta on CAD-based 6DoF visual tracking.

Takeaways: Going deep on a problem. The fundamentals of geometry and optimization used every day.