Join our Mechanistic Interpretability Group on LinkedIn

Cole McIntosh

AI and Data Engineer

Email Phone
Cole McIntosh

Stay updated on the latest in AI with my newsletter.

About Me

I am an innovative AI and Data Engineer based in Denver, Colorado, specializing in developing cutting-edge AI systems and data solutions. With a deep expertise in Large Language Models (LLMs) and AI Agents, I create sophisticated, scalable systems that push the boundaries of what's possible in artificial intelligence and data processing.

My work focuses on leveraging the power of LLMs to build intelligent, autonomous systems capable of complex decision-making and problem-solving. Using frameworks like Langchain, I develop agentic AI systems that can perform end-to-end automation of intricate business processes, significantly enhancing efficiency and unlocking new capabilities.

I have a proven track record of architecting and implementing AI-driven knowledge systems, multimodal AI solutions, and advanced data pipelines. My expertise extends to developing background worker AI agents capable of handling entire tasks autonomously, such as customer support. I also specialize in creating robust ETL solutions and building data-driven web applications. By combining technical prowess with a strategic mindset, I deliver AI and data solutions that not only solve current challenges but also pave the way for future innovations.

Experience

Pay Ready Logo

Data Engineer

Pay Ready (2023 - Present)

  • Developed agentic AI systems using Langchain for end-to-end automation of complex business processes.
  • Engineered AI-driven knowledge systems, enhancing information access and assistant evaluation company-wide.
  • Architected end-to-end LLM solutions for automating business workflows and analytics.
  • Implemented multimodal AI systems to process and analyze diverse data types, improving decision-making capabilities.
  • Created a RAG (Retrieval-Augmented Generation) Slack bot operating over internal documentation, significantly improving information accessibility and team productivity.
Otter Logo

Scaled Tooling Analyst

Otter (2022 - 2023)

  • Built scalable Python data pipelines for cross-team analytics and reporting, incorporating machine learning models for predictive insights.
  • Implemented AI-driven internal tools using natural language processing techniques to enhance operational efficiency and data-driven insights.
  • Developed a RAG (Retrieval-Augmented Generation) chatbot using Langchain, integrating Jira team spaces to enhance cross-team collaboration.
RealPage Logo

Business Analyst

RealPage (2021-2022)

  • Designed and maintained comprehensive business intelligence reporting systems.
  • Developed SQL data models for advanced analytics and automated client reporting.
  • Created automation pipelines to streamline business processes, significantly reducing manual workload.

Projects

Chain of Thought via Entropy Injection

Developed an innovative approach to Chain of Thought (CoT) reasoning using entropy-based injection with Llama 3.2 1B. The system dynamically injects CoT prompts based on output entropy thresholds, improving model reasoning capabilities while maintaining efficiency. Implemented with 8-bit quantization for memory optimization and features advanced token sampling techniques.

Llama 3.2 1B Mango

Developed a fine-tuned version of the Llama 3.2 1B model, optimized for chain-of-thought reasoning. This model was trained 2x faster using Unsloth and Hugging Face's TRL library, incorporating techniques such as LoRA, QLoRA, and RoPE scaling. The model was trained on the SkunkworksAI/reasoning-0.01 dataset, containing 29.9k examples to improve step-by-step problem-solving abilities.

Reranked RAG over My Personal Website

Developed a Streamlit app showcasing a Retrieval-Augmented Generation (RAG) system with reranking, designed to answer questions about me and my work based on my personal website content. Utilizes web scraping, LangChain, Cohere Rerank, Groq's LLM, and FAISS for efficient information retrieval and question answering.

Finance Fine Tuned Mistral 7B

Developed a fine-tuned version of Mistral 7B, specifically tailored for finance-related tasks. This model was trained on the alpaca finance dataset, enhancing its capabilities in financial analysis, prediction, and domain-specific language understanding.

Smol Vision

Built a local LLM vision pipeline for efficient image analysis without GPU requirements, demonstrating expertise in optimizing AI models for resource-constrained environments.

Structured Output with Multimodal Agents

Developed a system for generating structured output using multimodal agents, showcasing advanced AI integration and task routing capabilities.

Chain of Thought using Structured Output

Developed a front-end project demonstrating Chain of Thought reasoning using structured outputs with Mistral's Ministral 3B. The application streams different responses, allowing users to visualize the reasoning process in real-time. This project showcases advanced AI integration techniques and efficient use of smaller, open-source language models for complex reasoning tasks.

Skills

Python Langchain Machine Learning SQL Dev/ML Ops Natural Language Processing Data Pipelines PyCaret (AutoML) Huggingface LLM Evaluation Fine-Tuning RAG

Education & Certifications

Download CV (PDF)