A builder, researcher, and curious problem solver.
I’m a Computer Science student at the University of Texas at Dallas, and most of the work I enjoy sits somewhere between engineering depth and practical usefulness. I like building things that people can actually use, whether that means an AI workflow, a scalable data pipeline, or a hackathon prototype that turns into something memorable.
I’m especially interested in software engineering, AI systems, cloud infrastructure, and the way thoughtful product design can make technical work feel intuitive. A lot of my projects start with a simple question: how can I make this smarter, more useful, or more delightful?
At a glance
UT DallasB.S. in Computer Science, Expected May 2028, AES Scholar and Dean’s List.
LeadershipACM Research Lead, AKPsi Mu Rho Tech Cohort Lead, AKPsi VP of Operations.
Favorite buildsAI applications, product prototypes, and agentic systems with real-world use cases.
Work styleCurious, collaborative, and motivated by solving hard problems in a clean way.
What keeps me excited
Tackling technical challenges that have a visible outcome.
Building things from scratch and iterating fast.
Connecting research ideas to product-minded engineering.
Learning new tools when the problem calls for them.
Beyond the code
I wanted this site to feel personal, not just polished. Outside of engineering, I like basketball, competitive games like Valorant, and trying new food spots. I previously reached #34 in North America in Valorant. Those interests shape how I work too: I like fast feedback, friendly competition, and experiences that people actually enjoy.
The goal: technical enough to impress engineers, approachable enough to feel human.
Professional experience
I’ve had the chance to work across enterprise software, large-scale data systems, automation, and research. The common thread across all of it is building something reliable, useful, and grounded in real constraints.
What I focus on
Multimodal AIAgent securitySpark pipelinesCloud modernizationAutomationML research
IBM · Austin, TX
Software Engineering Intern — watsonx Orchestrate
Aug 2026 – Dec 2026
I work on bringing image inputs into watsonx Orchestrate, extending the information its agents can use beyond text. That also means handling how images fit into multimodal context windows and memory.
Preinvoke image inputs
Extended preinvoke plugins to accept image inputs, bringing image-based checks into the stage before an agent processes a request.
Conversation-aware plugins
Added conversation history across multiple chat-turn plugin calls. Preinvoke plugins can now use earlier messages as well as the current input, giving on-topic guardrails the context to assess whether a follow-up belongs in the conversation.
Multimodal support
Expanded support for image inputs while managing context-window and memory behavior. The work connects a new input type to the context an agent needs to respond.
Security validation
Built tools to detect prompt and code injection in image inputs. Image support introduces another place where instructions can enter a workflow, so validation is part of the feature itself.
Software Engineering Intern — Auto Loan Servicing Team
Jun 2026 – Aug 2026
I worked on moving legacy EMR workloads to AWS Glue for the Auto Loan Servicing team. The migration focused on modernizing Spark data processing and reducing cloud infrastructure costs while preserving the data downstream teams depend on.
Cloud migration
Worked on replacing legacy EMR cluster workflows with AWS Glue pipelines, moving the processing into a serverless ETL environment.
Data at scale
Engineered Spark batch jobs to process 7M+ records per run and deliver structured data to more than 20 teams.
Validation
Worked on back-testing the pipeline against two years of historical data to check output integrity before production deployment.
7M+ records per run · 20+ downstream teams · Two years of historical validation
AWS GlueApache SparkAWS EMRData pipelines
Alura · Frisco, TX
ML / AI Engineering Intern
May 2025 – Aug 2025
I built tools that connected content generation with day-to-day marketing work. The goal was to turn repeated manual steps into an autonomous agent workflow that could support multiple social accounts.
Agent architecture
Built and deployed an autonomous agent architecture integrated with ADK to automate marketing across 30+ accounts.
Workflow automation
Created content-generation and interaction tools and automated posting to social media platforms, saving more than six hours of manual work each week.
30+ accounts · 6+ hours saved per week
Google ADKAutomationAI agentsPython
ACM Research · Richardson, TX
Research Lead
Aug 2025 – Nov 2025
I led a five-person team studying how recurrent models can decode attempted speech from neural activity. We worked with more than 10,500 neural trials and focused on learning structure across different time scales.
Model design
Designed a Hierarchical Multiscale RNN with dynamic boundary detection. A Straight-Through Estimator let the model learn discrete boundaries as part of training.
Research focus
The architecture learns when to update representations at different time scales, rather than treating every time step identically.
Results
Improved word accuracy from 81.2% to 85.2% over the prior state-of-the-art RNN baseline described in the research.
Five-person team · 10,500+ neural trials · 81.2% → 85.2% word accuracy
PyTorchHM-RNNSpeech decodingResearch leadership
ACM Research · Richardson, TX
Research Intern
Sep 2024 – Nov 2024
I worked in a five-member team on a compact CNN–GNN hybrid model for EEG-based brain-computer interfaces. The project explored noninvasive neural signals and the potential for efficient models in neurological disorder management.
Exploration
Researched CNNs, LSTMs, GNNs, and t-SNE while testing on the PhysioNet EEG Dataset.
Hybrid modeling
Combined convolutional and graph-based modeling in a compact architecture, with edge-computing use in mind.
Results and recognition
Improved accuracy from 65.07% to 84.31%. The team received ACM Best Research in Fall 2024.
65.07% → 84.31% accuracy · ACM Best Research, Fall 2024
EEGCNNGNNEdge ML
Featured projects
These projects show the mix I enjoy most: practical engineering, thoughtful product decisions, and enough technical depth to make the build interesting.
Explore projects
Use the filters to focus the list, or hit surprise me for a random featured build.
Current pick: GlanceAR — a networking app that makes meeting people easier to remember and follow up on.
Mobile · AI · Networking
GlanceAR
React Native, Convex, MediaPipe, Meta AI
An AR networking app that supports opt-in facial recognition, gesture-based connections, and AI-generated conversation recaps so meeting people becomes easier to follow up on.
Python, Flask, JavaScript, YOLO, Gemini API, Google Cloud TTS
A real-time computer vision app for visually impaired users that detects objects, uses Gemini for context, overlays helpful cues, and announces alerts with audio.
A research-discovery and outreach platform that uses a RAG pipeline, web-search MCPs, and multi-agent email drafting to match students with labs more efficiently.
A full-stack AI-powered application that helps chapter members access information and connect more effectively, boosting reach-outs and increasing member usage.
+15% reach-outs300+ membersChat history
In progress · Recruiting intelligence
Vantage
Next.js, TypeScript, Python, Supabase PostgreSQL
A platform in development for finding relevant contacts connected to a job posting, helping applicants make connections around roles they’re interested in. The planned workflow brings job discovery and contact research together so each opportunity has a clear next step for outreach.
The full-stack skeleton and database foundation are implemented, including 16 tables, a metrics substrate, and a research queue. Ingestion, automated research, and outreach-draft workflows remain planned.
A regional customer-sentiment dashboard that connects a map of happiness scores with a multi-step diagnosis workflow. It retrieves recent snapshots, compares trends, identifies likely causes, and drafts internal and customer-facing messages.
The repository includes sentiment analysis with fallback behavior and a rule-based diagnosis workflow. It is a prototype, with simulated Slack publishing.
A team machine-learning project exploring binary classification of political statements from the LIAR dataset. The repository compares classical text-classification approaches with transformer-based models.
My public contribution added the BERT model. The project frames the dataset’s six veracity labels as a binary task; evaluation results are still in progress.
I like tools that help me build quickly, understand complex systems, and ship something real. Here are the buckets that best describe what I work with most.
