Mitchell Lamper
Specializing in generative AI, multi-modal diffusion models, and prompt optimization. From fine-tuning diffusion models to optimizing LLMs, to creating AI workflows, to testing and reinforcing models. I have experience in almost all facets of generative AI.
About Me
I'm a Machine Learning Engineer and AI Consultant based in New York, specializing in generative AI, multi-modal diffusion models, and prompt engineering. With a Master's in Data Science with a concentration in Machine Learning from Eastern University (4.0 GPA) and a background in Chemical Engineering, I bring both technical depth and creative problem-solving to AI challenges.
My most recent consulting contract was with Kroll Bond Rating Agency (KBRA), where I developed an internal prompt optimizer, built a prompt comparison tool, and evaluated the company's library of over 500 custom GPTs. Currently, I have an ongoing contract with Scale AI/Outlier as a Senior Prompt Engineer, where I craft evaluation benchmarks and train multi-modal models through RLHF frameworks, assessing them on instruction following, accuracy, and quality.
My expertise spans fine-tuning diffusion models (achieving 96% similarity scores on custom avatars), developing synthetic training data pipelines, and optimizing model parameters for peak performance. My experience across multiple AI startups has required versatility and self-reliance,independently identifying and resolving technical challenges across the full machine learning lifecycle, from dataset creation to model fine-tuning and inference optimization.
Generative AI
ML Frameworks
LLMs & Prompting
Cloud & DevOps
Languages & Data
Computer Vision
Featured Projects
Auto Prompt Optimizer Workflow
Developed an intelligent prompt optimization system at Kroll Bond Rating Agency focused on clarity, conciseness, hallucination risk, and determinism. Engineered a Prompt Comparison Tool that decomposes and cross-references instructions between prompts.
Custom Avatar Generation System
Achieved 96% similarity scores on personalized avatars using fine-tuned diffusion LoRAs at Mood Magic. Continually optimized learning rate, alpha, dimensions, and batch sizes. Implemented IP Adapter and Adetalier for enhanced consistency.
Synthetic Training Data Pipeline
Developed an automatic prompt creation pipeline using LLMs to generate synthetic training data for fine-tuning diffusion models. Created a Similarity Score tool to evaluate generated outputs and ensure quality consistency.
Multi-Modal LLM Training & Evaluation
Crafted unique prompts similar to HumanEval-Python and BigCodeBench to induce loss patterns in multi-modal models. Used RLHF framework to correct and reinforce models, assessing on instruction following, truthfulness, and quality.
Automated Image Captioning & Regularization
Implemented automatic image captioning for training data at Wand. Enhanced regularization by leveraging multi-modal LLMs to determine proper regularization classes and scraped images accordingly. Built synthetic prompt generation pipeline.
CNN Image Classifier for Architecture
Built a convolutional neural network for multi-class image classification of room and building types using TensorFlow and NumPy. Achieved high accuracy through careful architecture design and data preprocessing.
Get In Touch
Interested in collaborating on AI/ML projects or discussing prompt engineering? Let's connect!
Based in New York, NY