About
I design and build web applications end-to-end: React frontends, Go/Python backends, and data services that integrate with Oracle and Salesforce. I enjoy turning messy systems into clean APIs and dashboards that people actually use.
Featured Projects
Class Search Portal & Hub
React + Go middleware over vendor GraphQL/REST. Card-based results, filters that scroll with the cursor. Serving 40k+ users.
SuperTuxKart AI
Built CNN/Transformer models for classification, segmentation, detection; trained an autonomous driving agent.
Autoregressive Image Generator
Decoder-only Transformer that generates images token-by-token from a quantized codebook (BSQ/VQ). Trained end-to-end on SuperTuxKart frames.
More Projects
Inventory System
Track IT assets for CS department; REST API + MongoDB; deployed on Heroku.
Trip Mobile
Android travel app with Google Maps and Firebase realtime DB.
Flotsam
Arcade adventure built in Unity3D.
Job Bot
Automates job applications on multiple boards using Selenium.
Imgur Roulette
Flask app for random image discovery.
Azure AD Scripts
Automation for user & group provisioning with Python and PowerShell.
Flappy Bird VR
VR re‑imagining of a classic using Unity.
Save Emails
Tkinter app that logs your recent Outlook sent messages.
Disk Usage Scan
Win scripts to audit and trigger cleanups when storage > threshold.
UT Austin • AI/ML Coursework
Deep Learning
- Image Classification: Linear, MLP, deep residual networks on SuperTuxKart (80%+ accuracy).
- Semantic Segmentation & Depth Estimation: U-Net-style encoder-decoder with skip connections (75+ IoU, <0.05 MAE).
- Autonomous Driving Agent: MLP/Transformer/CNN planners predicting waypoints from lane boundaries or raw images.
- Custom training loops, TensorBoard logging, data augmentation pipelines.
Advances in Deep Learning
- Model Compression: 4-bit quantization, half-precision (FP16), LoRA adapters, QLoRA (7x memory reduction).
- Autoregressive Image Generation: Binary Spherical Quantization (BSQ) + decoder-only Transformer for token-by-token image generation.
- LLM Fine-tuning (SmolLM2): In-context learning (ICL), supervised fine-tuning (SFT), rejection sampling fine-tuning (RFT) for unit conversions.
- Vision-Language Models: Fine-tuned VLM and CLIP on SuperTuxKart for multimodal Q&A (70%+ accuracy).
Natural Language Processing
- Classical ML: Perceptron & Logistic Regression from scratch with feature engineering (77%+ accuracy on sentiment classification).
- Deep Averaging Networks: GloVe embeddings + feedforward nets; robustness to typos via prefix embeddings (74%+ on corrupted text).
- Transformers from Scratch: Self-attention, positional encodings, causal masking for character-level language modeling (6.3 perplexity).
- Fact-Checking: DeBERTa textual entailment to verify ChatGPT outputs against Wikipedia (83%+ accuracy).
- Fine-tuning ELECTRA: SNLI (89% accuracy) and SQuAD (78 EM, 86 F1).
Experience
Programmer Analyst • Pasadena City College
- Led React + GraphQL portal used by 40k+ students and staff.
- Designed Go middleware for vendor APIs; improved throughput and reliability.
- Streamlined Oracle data workflows; measurable performance gains.
Software Developer • Cloudsquare
- Built custom Salesforce apps (Apex, LWC, Visualforce).
- Integrated Microsoft Graph, CLEAR, Plaid, Experian, Equifax.
- Prototype: Raspberry Pi + TPU face detection system.
Database Administrator • ADR Services, Inc.
- Improved DB performance by ~60% with indexing and query tuning.
- Governance policies; BI reporting & data extraction.
Junior System Administrator • CSU San Bernardino
- Linux/Unix admin; automation via Ansible and Docker.
- Built Node/Express/Mongo tools for lab operations.