Skip to content
View TroyDanielsProjects's full-sized avatar

Block or report TroyDanielsProjects

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.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

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

Report abuse

Hi there, I'm Troy Daniels! πŸ‘‹

πŸŽ“ About Me

I am a Master's student in Computer Science at Columbia University , specializing in Machine Learning . My focus lies at the intersection of Mechanistic Interpretability, Generative Modeling, and MLOps.

I am passionate about building robust, interpretable AI systems and optimizing model deployment on cloud infrastructure.


πŸ”¬ Research & Projects

I am currently working on advanced research involving LLMs and Diffusion Models:

  • Uncertainty Quantification in LLMs (IBM & Columbia): Utilizing mechanistic interpretability to isolate "entropy neurons" in Llama 3.1 8B, successfully capturing 90% of incorrect generations via activation thresholding .
  • Manifold-Aware Diffusion Models: Developing extensions to DDPMs using anisotropic noise injection and local manifold estimation (k-NN/PCA) to improve sample complexity on non-Euclidean datasets .
  • Efficient Fine-Tuning: Implemented a pipeline combining GRPO, LoRA, and 4-bit quantization to fine-tune 7B parameter models on single-GPU hardware .
  • Medical VQA: Created a multimodal contrastive learning system for radiology image-question pairs, utilizing noise injection to improve robustness by 8.8% .

πŸ’Ό Professional Experience

Machine Learning / MLOps Intern | LTIMindTree

  • Enhanced large-scale RAG frameworks by extending AWS Bedrock deployments to support multimodal inputs .
  • Deployed computer vision systems for container ID detection, cutting deployment costs by 80% .

Software Engineer | The Hartford

  • Prototyped 3 LLM-based applications for document analysis, improving claims processing efficiency by 40% .
  • Delivered 30+ features for insurance underwriting systems in Java and implemented CI/CD pipelines .

πŸ› οΈ Skills & Technologies

  • Languages: Python, C++, Java, SQL
  • ML Frameworks: PyTorch, TensorFlow, Hugging Face, CUDA
  • Cloud & MLOps: AWS (Solutions Architect Associate), Kubernetes, AWS CDK
  • Domains: Computer Vision, NLP, Mechanistic Interpretability, Diffusion Models

🌐 Connect with Me

Pinned Loading

  1. AMLC-Final-Project AMLC-Final-Project Public

    Python

  2. NNDL-Final-Project NNDL-Final-Project Public

    Jupyter Notebook

  3. Infernaught/HPML_Final_Project Infernaught/HPML_Final_Project Public

    Python 1

  4. AIChessProject AIChessProject Public

    Implemented chess (No GUI) with a AI players

    Java

  5. MultiObjectTracking MultiObjectTracking Public

    My undergraduate thesis. Creating a multi-object tracking software with ML

    PureBasic