Below is an updated version of your professional overview, incorporating the latest details from your new resume. Feel free to adjust the content and formatting as you see fit.

Hello! I’m Jonathan Koch, a driven Software Engineer and AI Researcher with broad experience in both academic and industry settings. I’m passionate about Robotics, AI, and leveraging cutting-edge techniques—like LLM fine-tuning, contrastive learning, and multi-modal data processing—to tackle real-world challenges. My work spans from developing advanced object manipulation algorithms to designing high-impact ML solutions for image/video enhancement and job-search engines. Outside of tech, you can find me playing acoustic guitar, piano, or training in Brazilian Jiu-Jitsu.

• Languages: Python, C/C++, DJ (Diminished Java), C#, Java, RISC-V, Julia, Lua, JavaScript, Rust, Perl, R, CUDA
• Competencies: ROS, Actor-Critic, PyTorch, NumPy, Pandas, ONNX, CoreML, Linux, Networking, 2D & 3D Physics/Simulation Engines, LLM Fine-Tuning, Prompt Engineering, Multi-Modal Data Processing, Semantic & Generative Models, Self-Supervised Learning, Compilers
• Research Background: Robotics (Object Manipulation, Perception Modeling, Reinforcement Learning, Model Predictive Control, Reward Tuning, Variational Autoencoders), NLP (Semantic Search, Text-Based Recommendation Systems, Multi-Modal Search), Audio & Image Processing (Quality Assessment, Enhancement, Segmentation)
• Interests: Predictive Models, Multi-Agent Learning, Underactuated Robotics, Self-Supervised Learning, Cognitive Science, Data Science, MIT OpenCourseWare, Acoustic Guitar, Piano, BJJ
• Personal Website: Jonathanzkoch.dev

Bachelor of Science in Computer Science, Concentration in Robotics and AI
University of South Florida College of Engineering (Fall 2020 – Spring 2024)

• Engineering GPA: 3.7/4.0

November 2023 – Present | Tampa, FL

• Enhanced core value proposition of a job board through data-driven matching, streamlined processes, and scalable ML pipelines.
• Contrastively fine-tuned vector search models on real and synthetic user data (job candidates) to improve relevance and discoverability of job postings, leveraged Matroskya representation learning to get low dimensional vector search and keeping comparison FLOPs as low as possible for speed.
• Leveraged LLM agents to scrape over 30k jobs daily from 12k+ career pages and boards; automated application workflows for users, boosting premium revenue and user engagement.
• Reduced LLM inference costs and enhanced pipeline efficiency, cutting scraping costs by 43% and speeding daily scrapes by 2.76×.
• Delivering advanced ML features to serve 20k+ monthly active users in pursuit of the ideal career match.

June 2024 – Present | Addison, TX

• Designed, trained, and integrated ML models for quality assessment, segmentation, semantic analysis, and enhancement of images/videos.
• Architected a zero-click enhancement pipeline for cloud API users, allowing dynamic multi-step processing and incremental quality checks.
• Built internal tools for large-scale data analysis, checkpoint comparisons, and streamlined R&D workflows.
• Implemented user data tracking to develop suggestion and autopilot tools, reducing average clicks per export session from 21.8 to 3.6.
• Led iOS app ML integration (CoreML) for 100k+ monthly users, debugging edge cases across multiple iPhone hardware versions.
• Deployed automated conversion-failure detection using a high-precision classification pipeline to maintain model quality across ONNX, CoreML, TensorRT, and OpenVINO conversions.

Fall 2021 – Present | Tampa, FL

• Investigated robotic object manipulation with supervised and reinforcement learning, focusing on transformer-based perception and policy optimization.
• Explored graph neural networks and transformers to model dynamic relationships between states and actions.
• Leveraged contrastive divergence in latent space conditioning on actions to improve predictive accuracy and representation quality.
• Developed heuristic approaches based on physics-inspired predictive models, enabling more robust grasping strategies without reliance on large-scale RL data.

November 2023 – May 2024 | Tampa, FL

• Prototyped a Generative Agents architecture using large language models capable of reasoning and explainable action steering in simulations.
• Deployed contrastive learning for text classification, boosting accuracy from 93% to 99%.
• Created training tools to mitigate overfitting by generating structured but semantically varied data to strengthen model robustness.

May 2022 – November 2023 | Tampa, FL

• Developed Windows and Linux lab nodes for parallel simulation hardware, collaborating with embedded teams to process sensor data in real-time.
• Integrated repositories into CI/CD pipelines and tackled a wide range of R&D projects, including multi-screen touchscreen driver development and advanced MLOps integrations.

• IEEE Open-Source Robotics Software Stack
  GitHub

  • Facilitated autonomous capabilities (perception, SLAM, and planning) with distributed computation across multiple devices.
  • Modeled environmental dynamics and latent state representations for RL and behavior cloning on Jetson hardware.

• Teach-A-Bull (AI Tutor)
  GitHub | Competition Video

  • Leveraged LLMs to generate multi-layered educational content (lessons, homework, schedules).
  • Achieved a 10× decrease in SAT/ACT prep costs and a high similarity to expert-curated content.

• MicroGradPlus
  GitHub

  • A lightweight computational graph / auto-differentiation API using only NumPy, inspired by Karpathy’s Micrograd.
  • Achieved 94% accuracy on MNIST and perfectly matched PyTorch gradients on vector tests.

• CoderSchoolAI
  GitHub | Demo

  • A beginner-friendly neural network and AI tools library for educational purposes.
  • Simplified ML modeling for novice learners and kids by providing black-box design for data-driven tasks.

• Virtual Assistant
  GitHub

  • An NLP-based assistant for task sequencing.
  • Processes natural language commands to actionable steps.

• Email: [email protected]
• Phone: (631) 372-1384
• Location: Addison, TX
• LinkedIn / Personal Site: jonathanzkoch.dev

I take pride in creating robust, elegant solutions—whether it’s designing complex robotics systems or building user-centric AI experiences. My curiosity drives me to constantly learn, refine my craft, and push the boundaries of what AI can do.