Some of these projects were completed as a freelance consultant for companies, while others are personal research initiatives. Only the most relevant projects are listed here.
LLM systems
LLMs for Smart Contract Vulnerability Detection
As the first ML hire at a blockchain security startup, I led a multi-stage pipeline with over 20 models to detect, explain, and fix vulnerabilities in smart contracts. False positives reduced from 50% to less than 10%.
LLM transformer
A transformer-based code completion system focused on Kotlin, flexible enough to be trained on other languages. Optimized to train and run on a single NVIDIA RTX 4090 GPU.
GitHub →Generative AI
Generative AI for Personalized Face Image Creation and Modification
Rapid LoRA-based finetuning of Stable Diffusion that personalizes a model from a few photos in under one minute, and lightweight GANs to modify specific facial features.
Stable Diffusion
An educational implementation of diffusion models, designed to be trained from scratch on a single NVIDIA RTX 4090 GPU. Built with PyTorch and containerized with Docker.
GitHub →Edge AI
Embedded Quantized Models
Two edge AI projects deploying quantized deep learning models for real-time applications: a lightweight face embedding model for person identification on resource-constrained devices, and an image classification system to detect adversarial or manipulated images.
3D and geometry
HashNeRF
A PyTorch implementation of the Hash NeRF architecture, using multi-resolution hash encoding for fast and memory-efficient 3D scene representation. Data preprocessing, training and rendering are containerized with Docker.
GitHub →3D Point Cloud Instance Segmentation
A prototype for instance segmentation of large-scale 3D point clouds generated by laser scanning of indoor environments, classifying points into semantic objects and grouping them into instances.
3D Virtual World Generation from Text
A 5-month R&D effort to train a reinforcement learning agent that operates Houdini's GUI directly to construct 3D scenes based on text prompts.
Predictive modeling
YCombinator Startup Success Prediction
Predictive machine learning models that forecast the success of Y Combinator startups based on founder attributes and characteristics, analyzing historical data from YC batches to provide data-driven insights for early-stage investment decisions.