Projects

Biochar-concrete company turning food waste into lower-emission concrete with measurable carbon outcomes.

• 63,550+ lb concrete poured across pilot deployments
• 20,500+ lb CO2 sequestered and 24,600+ lb food waste diverted
• Ran iterative material-testing workflows across pilot batches to improve durability and consistency
• Accepted into Launch Chapel Hill and secured $xk+ in funding

PFAS and nitrate water-filtration systems using regenerable ion exchange for underserved communities.

• 16+ active deployments across 3 countries with 1,230,000+ L/year treated water throughput
• Designed regenerable ion-exchange columns with breakthrough-curve monitoring
• Built field telemetry and maintenance scheduling workflows across multi-country deployments
• Engineered systems for strong contaminant removal at low deployment cost

Built reward-flight recommendation systems and backend APIs at a Y Combinator-backed travel fintech platform.

• Built ranking features for route selection and reward-value prioritization
• Integrated ML outputs into Node.js + PostgreSQL production services
• Added evaluation instrumentation for latency, hit-rate, and recommendation quality
• Contributed to systems used by over 20,000 current Chrome extension users

Research with Duke NRFI on how climate and policy signals affect green-bond pricing and infrastructure deployment economics.

• Built an end-to-end Python + R pricing pipeline that links bond trades, issuer data, and macro indicators
• Integrated Bloomberg, TRACE, and FRED into a clean bond-level panel with climate and policy features
• Applied matched-pair, fixed-effects, and event-study models to estimate greenium drivers in plain terms
• Ran VAR/GARCH shock analysis to show how policy and climate-risk events reprice bond spreads over time

ASL learning platform with real-time gesture recognition and instant feedback using computer vision.

• Engineered real-time CV pipeline for hand-landmark extraction, normalization, and temporal smoothing
• Built TensorFlow.js inference stack with calibrated confidence thresholds and low-latency browser inference
• Implemented feature-windowing and sequence-aware gesture detection to reduce false positives in live sessions
• Added per-sign accuracy tracking, feedback scoring, and progression analytics for adaptive practice
• Won 1st in the Congressional App Challenge

AI-powered satellite risk analysis system for rapid post-disaster damage assessment and relief prioritization.

• Built CNN-based damage-severity classification for remote-sensing imagery
• Implemented image preprocessing and tiling pipeline for large-area satellite scans
• Designed risk-priority scoring outputs to support faster triage in disaster response workflows
• Developed and shipped as a hackathon prototype with full technical demo

Internship focused on AI-powered music personalization using multimodal context and emotion cues.

• Implemented facial-expression classification pipeline for emotion-conditioned recommendations
• Integrated text/voice context features into ranking logic for playlist generation
• Shipped low-latency API flows connecting ML inference outputs to end-user recommendations