Portfolio of Work · Class of 2026
Generative AI, applied to business.
I'm Asarel (Tito) Morales. In 2026 I completed the Post Graduate Program in Generative AI for Business Applications at the McCombs School of Business, The University of Texas at Austin — five months of hands-on labs that run the whole arc: from regression and neural networks, through transformers and LLMs, to retrieval-augmented chatbots, fine-tuning, agentic AI, and multimodal generation. This site is the record of that work.
The Journey
The curriculum was sequential by design — each module built on the last. Every chip below is a lab notebook I built and ran, February through June 2026.
Foundations: Python & Data Science
Python for data work — pandas, NumPy, visualization — and the statistical groundwork every model after this depends on.
- DataScience
- LinearRegression
- PredictionModel
Machine Learning for Business Problems
Classification and regression applied to real business cases: fraud, risk pricing, public-health data, recommendations.
- CreditCardFraudDetection
- MedicaidFraud
- InsurancePremiumPrediction
- COVID-19
- MovieRecommendationSystem
Deep Learning & Neural Networks
From a single perceptron to trained networks — building digit classifiers twice, once with classical ML and once with deep learning, to feel the difference.
- NeuralNetwork
- Classifier · ml_digit_classifier
- Classifier · dl_digit_classifier
NLP, Embeddings & Sentiment
Text as data: tokenization, embeddings, and sentiment analysis on product reviews and airline customer feedback.
- Sentiment
- ProductReview
- AirlineCustomerSentimentAnalysis
- Embeddings.Transformers
Transformers & Text Generation
Inside the architecture that powers modern AI — attention, decoding, and generating text with transformer models.
- TransformersforTextGeneration
LLMs & Prompt Engineering
Working with large language models deliberately: prompt patterns, structured outputs, and applying them to workflows like support-ticket triage.
- LLM.Prompt.Engineering
- SupportTicket
- SmartResearchAssistant
- PoweredResearchAssistant
RAG & Enterprise Chatbots
Retrieval-augmented generation: grounding LLMs in a company's own data. Built order-query, restaurant, healthcare-audit, and medical-assistant bots.
- KartifyOrderQueryChatBot
- FoodHubChatbot
- RestaurantRatings
- HealthcareAuditChatbot
- MedicalAssistant
- MedicalDiagnosis
Fine-Tuning LLMs
Adapting foundation models to a task: hands-on fine-tuning and an automated quality-classification model trained on domain data.
- HandsOnFineTuning
- AutomatedQualityClassification (Fine-Tuned)
Agentic AI
From chatbots to agents — models that plan, call tools, and act. Capped with a personalized news-discovery agent.
- Agentic.AI.Intro
- AIPoweredPersonalizedNewsDiscoveryAgent
Responsible AI, Security & Ops
Shipping AI like it matters: responsible-AI practice and security, plus the MLOps and LLMOps that keep models alive in production.
- Responsibility-AI-Security
- Introduction-to-MLOps
- Introduction-to-LLMOps
Multimodal Generative AI — Capstone Case Study
The Browbake case study: one brief, three modalities. Generated a full marketing campaign — copy, imagery, and audio — from a single pipeline.
- BrowbakeCaseStudy
- MultiModal · campaign pipeline
Featured Work
A closer look at the builds that best show the range.
Browbake: a marketing campaign generated end-to-end
For a fictional bakery brand, one pipeline produced the whole campaign: LLM-written copy, diffusion-generated product imagery (with classifier-free guidance), and text-to-speech audio spots. Iterated across multiple notebook versions to swap TTS engines and tune generation quality.
Healthcare Audit Chatbot
A retrieval-augmented assistant that answers audit questions grounded in healthcare policy documents — an LLM that cites its sources instead of guessing.
Automated Quality Classification
Fine-tuned a foundation model into a domain-specific quality classifier — the "small model, your data" pattern businesses actually deploy.
Personalized News Discovery Agent
An agent that plans, fetches, filters, and summarizes news to a user's interests — tool-calling and autonomy beyond a single prompt.
Stock Intelligence Suite
Outside the curriculum I applied each week's techniques to markets: predictors (XGBoost, classical ML), sentiment-augmented daily forecasts, a signal bot, and a market-data agent.
Toolbox
What I now reach for, grouped the way the work actually happens.
Data & ML
- Python · pandas · NumPy
- scikit-learn · XGBoost
- Neural networks (Keras/TensorFlow)
- Google Colab notebooks
Generative AI
- Transformers & embeddings
- Prompt engineering
- Retrieval-augmented generation
- LLM fine-tuning
Multimodal
- Diffusion image generation
- Text-to-speech pipelines
- ComfyUI workflows
- Local video generation (LTX-Video)
Shipping It
- Agentic AI & tool use
- MLOps / LLMOps foundations
- Responsible AI & security
- Hugging Face ecosystem
Beyond the Program
The certificate was a starting line, not a finish line.
PG Program in AI Agents for Business Applications
Enrolled in the follow-on McCombs / UT Austin program — going deeper on agentic systems, orchestration, and automation for real business workflows.
Local AI experiments
Running generative pipelines on my own hardware: ComfyUI image workflows, LTX-Video, and self-hosted tooling on a personal VPS — this site included.
Markets as a testbed
An ongoing series of stock-prediction and sentiment projects — my way of pressure-testing every new technique against noisy, unforgiving data.