Skip to content
View marcovizcarra's full-sized avatar

Block or report marcovizcarra

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
marcovizcarra/README.md

Marco Vizcarra

Stanford M.S. Electrical Engineering · AI/ML · Reinforcement Learning · Robotics

Engineer working across reinforcement learning, multimodal learning, computer vision, and robotics, with 14+ years of industry experience building software, hardware, automation, and complex product systems at Apple and Skyworks.

Portfolio · Projects · LinkedIn


Selected Research & Projects

Stanford CS224R · Deep Reinforcement Learning · Honorable Mention

Developed a frontier curriculum for RLOO fine-tuning of Qwen2.5-0.5B, improving pass@1 from 0.41 → 0.49 over vanilla RLOO.

Case Study →

TidyBot2: Language-Guided Mobile Manipulation

Stanford CS339R / ME326 · Collaborative Robotics

Built the ROS2 navigation and sim-to-real motion layer for a TidyBot2 mobile manipulation system using MuJoCo and real robot hardware.

Case Study →

Data Augmentation for Offline RL

Stanford CS234 · Reinforcement Learning

Evaluated Behavior Cloning, IQL, and CQL under visual distribution shift across 500 unseen CoinRun levels, studying how augmentation affects offline-RL generalization.

Case Study →

Stanford CS231N · Computer Vision

Fine-tuned SigLIP SO400M-384 for fine-grained recognition across 555 NABirds species, reaching 83.78% Top-1. The team's SigLIP + ConvNeXt late-fusion system reached 92.48% Top-1.

Case Study →


Technical Focus

Python · PyTorch · C/C++ · ROS2 · MuJoCo · Linux · Git

Reinforcement Learning · Robot Learning · LLM Fine-Tuning · Multimodal Learning · Computer Vision

marcovizcarra.com →

Pinned Loading

  1. frontier-curriculum-rloo frontier-curriculum-rloo Public

    Frontier-curriculum RLOO for LLM reasoning, developed for Stanford CS224R. Includes baseline comparisons, a simplified public implementation, evaluation, and results. Improved pass@1 from 0.41 to 0…

    Jupyter Notebook

  2. fine-grained-bird-recognition fine-grained-bird-recognition Public

    Fine-grained bird recognition on NABirds using part-aware ConvNeXt, SigLIP and late logit fusion.

    Python

  3. procgen-offline-rl procgen-offline-rl Public

    Offline RL on Procgen CoinRun with visual data augmentation, comparing Behavior Cloning, IQL and CQL for generalization to unseen levels.

    Python