Member of Technical Staff
Stealth AI Startup
Dec 2024 — PresentBuilt core diffusion and VLM systems as the startup’s second technical hire, with reported gains in model quality, annotation cost, and training and inference speed.
• Second technical hire and primary contributor across model architecture, training code, and captioning infrastructure, independently implementing most of the end-to-end development pipeline.
• Designed the proprietary Diffusion Transformer (DiT) architecture and training methodology, achieving quality competitive with models up to 5× larger in parameter count in internal human preference evaluations.
Computer Vision R&D Engineer
Ik2 Sàrl / SkyCues
Feb 2024 — Mar 2025Developed satellite super-resolution models for production mapping, converting 10 m imagery into outputs at 62.5 cm pixel spacing.
• Developed Transformer-based satellite super-resolution models with ×2, ×4, ×8, and ×16 upscaling, generating outputs at 62.5 cm pixel spacing from 10 m imagery for production mapping.
• Designed losses and preprocessing for sensor noise, aliasing, and misalignment; built models that generalize across satellite vendors and imaging conditions.
Machine Learning Engineer
Government Technology Organization
Jul 2023 — Jan 2024Built ML models and supporting data and evaluation pipelines for internal government use cases; project details are under NDA.
• Developed ML models, data preprocessing, and evaluation pipelines for internal operational use cases. Project details under NDA.
Computer Vision Fellow
Fellowship.ai
Apr 2023 — Jun 2023Developed diffusion-based virtual try-on and image synthesis models focused on preserving garment details, texture, and pose.
• Developed diffusion-based virtual try-on models and image synthesis pipelines to preserve garment details, textures, and target pose.
Machine Learning Engineer
The Cramer Project
Nov 2022 — Jun 2023Built and deployed OCR for handwritten Kyrgyz-language documents, reaching 99.32% character recognition accuracy on internal benchmarks.
• Developed and deployed an OCR system for handwritten Kyrgyz-language documents, achieving 99.32% character recognition accuracy on internal benchmarks.
