Georgia Tech (BS) CS + EE
currently focused on inference, ML systems, and hardware-software co-design.
A collection of GPT-style transformer implementations and experiments in PyTorch and JAX.
The PyTorch model references Andrej Karpathy's tutorial and generates Shakespearean text. The JAX model is trained on OpenWebText and was built after working through Attention Is All You Need and DeepMind's Scaling Book.
Tech Stack: PyTorch · JAX · Flax NNX · Optax · Orbax · XLA · Python
Highlights:
- Implemented causal self-attention, multi-head attention, residual blocks, layer normalization, and autoregressive generation
- Built next-token prediction training loops in both PyTorch and JAX
- Used
jax.jit, automatic differentiation, Optax, and XLA for accelerated training - Continuing to add new transformer architectures and experiments
Links:
- 🔗 Repo
What I Built
- PyTorch GPT-style model trained to generate Shakespearean text
- JAX GPT-style model trained on OpenWebText
- Token and positional embeddings
- Causal multi-head self-attention
- Feed-forward neural networks
- Pre-norm residual transformer blocks
- Next-token cross-entropy training
- Autoregressive generation with temperature and top-k sampling
- Checkpointing and accelerated training with Orbax and XLA
AI crop-monitoring platform that detects plant disease, nutrient deficiencies, and hydration issues from images, then generates personalized plant-health reports.
Tech Stack: React · Flask · Python · TensorFlow · CNNs · LLM Workflows · Tailwind
Highlights:
- Reached 86% CNN accuracy after model tuning
- Combined image classification with an LLM workflow to generate personalized recommendations
- Placed 2nd internationally at TSA Software Development
- Turned down a $10,000 investment offer
Links:
- 🔗 Repo
Key Features
- Plant disease detection from uploaded images
- Nutrient deficiency and hydration analysis
- LLM-generated health recommendations
- Farm health and operations dashboard
- Financial tracking
- Saved results and historical analysis
AI medical-bill coordination platform for law firms, healthcare providers, and funders.
Tech Stack: Next.js · TypeScript · Flask · Python · Claude API · Pydantic · CMS API · Medicare Data · GCP
Highlights:
- Extracts structured line items and billing data from uploaded medical documents
- Compares charges against CMS and Medicare reimbursement data
- Supports role-based workflows for law firms, providers, and funders
- Handles inconsistent document layouts while keeping token usage efficient
- Deployed with separate cloud-hosted frontend and backend services
Links:
- 🔗 Repo
Key Features
- Medical-bill upload and processing
- Line-item and total-charge extraction
- CMS and Medicare reference-data integration
- Role-based dashboards and permissions
- Structured JSON output from AI parsing
- Human-readable billing summaries
- Error handling for inconsistent bill formats
- Cloud deployment setup
AI language-learning platform that lets students scan real-world objects for Tamil translations, transliterations, audio, flashcards, and quizzes.
Tech Stack: Next.js · React · Flask · Gemini Vision · SQLAlchemy · JWT Auth · Tailwind
Highlights:
- Piloted with a local Tamil school serving 1,000+ students
- Uses computer vision to turn real-world objects into vocabulary lessons
- Supports classroom and at-home language practice
Links:
- 🔗 Repo
Key Features
- Object scanning for vocabulary learning
- AI-generated translations and transliterations
- Personal word bank
- Flashcards and quizzes
- Streak tracking
- Statistics and achievements
- Classroom and at-home practice
Official website for a local IT company, designed to present its services, credibility, and client work.
Tech Stack: React · TypeScript · Tailwind · Flask · Vite
Impact: Helped present services to 20+ clients
Links:
- 🌐 Website
What I Built
- Responsive landing page
- Service and company-overview sections
- Modern branded interface
- Client-facing web presence
- Backend contact integration
Website for my 501(c)(3) nonprofit expanding access to coding education through workshops and student-led programs.
Tech Stack: React · Tailwind · JavaScript
Impact: Supported coding programs serving 250+ students and contributing 2,600+ volunteer hours
Links:
- 🌐 Website
What I Built
- Nonprofit landing page
- Program and workshop information
- Student-facing resources
- Responsive interface
- Outreach-focused design
A lightweight daily planner built around my personal workflow for organizing tasks and goals.
Links:
- 🔗 Repo
|
|
|
|