Daniel Regalado Cardoso

AI for Equitable Public Transit

Deloitte Capstone · Jan 2026 – Present

Predictive models integrating Census ACS, GTFS, and ArcGIS data to identify transit equity gaps in Miami-Dade with SHAP-based interpretability.

Python SHAP GTFS ArcGIS

Repo · Demo

AI Scheduling Assistant

UHealth Bascom Palmer · Capstone · Sep 2025 – Present

Prototyped a LangGraph + GPT-4 multi-agent chatbot with RAG and real-time monitoring; shipped the production solution on Microsoft Copilot Studio for 100+ providers.

Copilot Studio LangGraph GPT-4 RAG

HDL Cholesterol Prediction

ASA Data Challenge — Top 10% · 2026

CatBoost model on NHANES data (R² = 0.75) with domain-guided feature engineering and SHAP analysis.

CatBoost SHAP NHANES

Repo

SQL Agent with LLMOps

Fine-Tuning + RAG · 2026

Multi-model SQL agent: fine-tuned text-to-SQL, chart reasoning, and SVG rendering via Unsloth and knowledge distillation, deployed on HuggingFace Spaces.

Unsloth Text-to-SQL RAG LLMOps

Repo · Demo

MoE Backward Optimization

Triton Kernels · 2026

Hand-derived backward passes and fused Triton kernels for Mixture-of-Experts training, following Unsloth’s kernel-fusion playbook on expert-routing layers.

Triton PyTorch MoE Kernel Optimization

Repo

ATS Resume Optimizer

Live on HuggingFace · 2026

Hybrid ML scoring (Llama 3.1-70B + Sentence-Transformers) that analyzes resumes against job descriptions and generates ATS-optimized rewrites.

Llama 3.1-70B Sentence-Transformers Gradio

Repo · Demo

MayaVoice LLM

GenAI for Mayan Languages · 2026

Multilingual GenAI for translation and conversation in Guatemalan Mayan languages, bridging low-resource indigenous languages with modern AI.

LLMs NLP Low-Resource Languages Gradio

Repo · Demo

DiagBoostTS

Python Library · 2025 – Present

Iterative diagnostic boosting for time series forecasting — systematic EDA → Stationarity → Model → Evaluate pipeline.

Python Time Series XGBoost SARIMA

Repo

Coffee Shop Big Data Analytics

PySpark + XGBoost · 2026

End-to-end big-data ML on ~1.8M coffee-shop transactions: PySpark for ingest and EDA, XGBoost + Optuna for modeling, SHAP for explanations. Wait time R² 0.34, purchase amount R² 0.86, rewards-member AUC 0.98.

PySpark XGBoost Optuna SHAP

Repo