Notebook

Technical deep-dives into systems I've built. Each project solves a specific problem.

Research & Quantum

Generating Quantum GHZ States in Rydberg Arrays with Optimal-Control Pulses

Achieved N=20 Schrödinger cat states with ≥0.54 fidelity using RedCRAB optimal-control pulses in Rydberg atom arrays. Characterized phase coherence via parity oscillations to understand decoherence limits in multi-qubit quantum systems.

Quantum ComputingPythonOptimal Control+2

View

Engineering Steel Microstructures with Computational Thermodynamics

Built computational tool using pycalphad for phase diagram generation and steel microstructure prediction. Designed Fe-C carbon steels and Fe-Cr-C martensitic stainless steels (Type 410) through binary and ternary phase diagram calculations. Implemented eutectic/eutectoid identification, phase fraction calculations via lever rule, and polynomial fitting for phase boundaries.

PythonpycalphadNumPy+3

View

DRAM Reliability Analysis Under Thermal and Mechanical Stress

Performed computational reliability analysis of DRAM devices under thermal and mechanical stress conditions. Built Python simulation framework to model temperature fields, stress distributions, and scaling trends in semiconductor memory. Generated visualization of geometry, temperature fields, stress fields, and device scaling trends for reliability assessment.

PythonNumPyMatplotlib+4

View

ML & AI

Building AI-Powered Media Kits for Content Creators Using TikTok API

Automated creator media kit generation by integrating TikTok API, Instagram Graph API, and GPT-4 to generate professional media kits in 30 seconds. Reduced kit creation time from 2-3 hours to 30 seconds with AI-powered bio generation and Puppeteer PDF rendering.

Next.jsTikTok APIInstagram API+5

View

Training Regression Model on 170k Songs for Real-Time DJ Mix Recommendations

Built regression model on AWS Sagemaker analyzing BPM, energy, valence across 170k tracks to predict 'mixing scores' between songs. Deployed real-time inference via Lambda + API Gateway, enabling DJs to find compatible transitions in <500ms. Integrated Spotify API for genre classification.

AWS SagemakerPythonML+4

View

Predicting Stock Movement with NLP Sentiment Analysis on News

Built pipeline scraping 10k+ financial news articles daily, running VADER sentiment analysis, and training regression model to predict next-day stock movement. Deployed on AWS Lambda with S3 storage, achieving 62% directional accuracy on S&P 500 stocks.

PythonNLPVADER+3

View

Optimizing Chip Layout with Reinforcement Learning for AMD

Applied deep RL (PPO) to optimize transistor placement in chip layouts for KPMG Ideation Challenge. Trained agent with custom reward function balancing wire length, congestion, timing, and thermal constraints. Achieved 12% improvement over manual layouts and 48x speedup.

PythonPyTorchReinforcement Learning+3

View

Generating Music Patterns with RNNs Trained on MIDI Data

Trained LSTM network on 5,000+ MIDI files to generate novel musical sequences with attention mechanism and tempo/key conditioning. Built real-time web interface for producers to generate 8-bar variations from seed melodies. Achieved 4.1/5 rating from professional producers with <500ms latency.

TensorFlowPythonLSTM+5

View

Detecting Microexpressions with CNN for Clinical Diagnosis

Built Congruence, a multimodal emotional AI platform using CNNs to detect microexpressions and voice stress for clinical diagnosis. Achieved 76% accuracy on 7-emotion classification, deployed to 48+ psychiatric clinics with real-time analysis. Reduced clinical documentation by 92% while improving diagnostic accuracy by 18%.

TensorFlowCNNPython+4

View

Full-Stack Apps

Automating Purchase Order Verification with OCR + Product Matching

Built full-stack PO processing system using Tesseract OCR + fuzzy matching to auto-verify line items against product catalogs. Reduced verification time from 15 min/order to <30 seconds (30x speedup). Achieved 93% automation with 99% accuracy using PostgreSQL full-text search and interactive verification UI.

ReactTypeScriptFlask+5

View

Scaling NIL Marketplace for College Athletes with Payment Processing

Built OpenField, a two-sided marketplace connecting 520+ college athletes with local businesses for NIL deals. Integrated Stripe Connect for split payments with 15% platform fee, automated contract generation via DocuSign, and NCAA compliance monitoring. Processed $52k+ in transactions with <48 hour deal completion time.

ReactNode.jsTypeScript+4

View

Scaling Social Fashion App to 300+ Beta Users on Mobile

Built FitCheck, a React Native social network for outfit sharing that scaled to 320+ beta users with 40% DAU. Optimized image compression reducing upload time 67% (15s→5s), implemented personalized feed algorithm with Redis caching (90% hit rate), and processed 10,400+ posts. Achieved 62% Day-7 retention with infinite scroll and real-time updates.

React NativeNode.jsMongoDB+4

View

Automating Lead Generation for Local Businesses with AI Agents

Built AI-powered lead generation system using LangChain agents that scrapes business directories, enriches data with ML lead scoring, and generates GPT-4 personalized emails. Processed 62,000+ prospects achieving 8.2% response rate (4x industry average). Delivered 300% growth for FitCheck, $2k+ revenue for Workwear, and 2x ROI for Gloss Authority across 15+ clients.

PythonLangChainOpenAI+5

View

Building Real-Time Investment Dashboard with Stock + Crypto Analytics

Unified portfolio tracker pulling live data from Yahoo Finance + CoinGecko APIs. Built Flask backend with Pandas for performance calculations (ROI, Sharpe ratio, volatility) and Plotly visualizations. Cached price data in MySQL reducing API calls 80%. Implemented MySQL stored procedures for VaR and max drawdown calculations.

FlaskMySQLReact+6

View

Search & Retrieval

Building Semantic Search for E-Commerce with Vector Embeddings

Implemented semantic product search using Pinecone vector DB + OpenAI embeddings to replace keyword-based search. Improved product discovery CTR by 40% by enabling natural language queries like 'sustainable winter jacket under $100'.

PineconeOpenAI EmbeddingsNext.js+5

View

Building Image Search Engine with CLIP Embeddings + FAISS

Indexed 100k+ images using CLIP embeddings and FAISS vector search, enabling semantic queries like 'sunset over mountains'. Deployed on AWS with <200ms latency. Implemented approximate nearest neighbor search for sub-linear query time.

CLIPFAISSPython+5

View

Data & Visualization

Analyzing 500+ EV Models with Interactive Tableau Dashboard

Cleaned and normalized dataset of 500+ electric vehicles, standardizing range/battery metrics across EPA, WLTP, and NEDC standards. Built interactive Tableau dashboard with dynamic filters for price, range, brand comparisons. Implemented Python outlier detection identifying 12% data quality issues. Revealed key insights: $45k/300mi sweet spot, Korean manufacturers best value.

TableauPythonPandas+5

View

Visualizing Material Properties with Interactive D3.js Charts

Built interactive Ashby-style scatter plot with D3.js visualizing mechanical properties (strength, stiffness, density, cost) for 50+ materials across metals, ceramics, polymers, and composites. Implemented log-scale axes, zoom/pan, category filtering, and hover tooltips. Calculates performance indices (strength-to-weight, specific modulus) for material selection. 1,000+ student views.

D3.jsJavaScriptSVG+4

View

Systems & Infrastructure

Building High-Performance Crypto Trading Engine in C++

Built high-frequency trading engine in C++ with sub-millisecond order execution (18.5μs latency), options strategies (covered calls, straddles), and backtesting. Achieved 54k orders/second throughput with StreamLit dashboard for real-time portfolio analytics and market data visualization.

C++17Multi-threadingSTL+4

View