resume

[education]

University of WaterlooExpected April 2026

Bachelor of Software Engineering (BSE)

Software Engineer InternJan 2026 – Present

Fish AudioSan Francisco, CA

-Built an agentic video creation pipeline in React/TypeScript with Zustand state management, orchestrating script generation, asset selection, and TTS audio synthesis into an auto-populated timeline.
-Designed a script editing interface using Next.js with drag-and-drop line reordering, per-character voice assignment, and selective regeneration, giving users granular control over AI-generated TTS before final render.
-Built the speech-to-text product page end-to-end in Next.js with file upload, live browser via WebRTC, and async transcription polling against a Replicate-hosted ASR model.

Software Engineer InternApr 2025 – Aug 2025

NationGraphSan Francisco, CA

-Built an entity resolution ML pipeline in PyTorch to normalize 600M+ vendor names, combining TF-IDF embeddings with custom supervised classification models to deduplicate and standardize records at scale.
-Built dynamic Python web scrapers using Selenium and BeautifulSoup to extract and normalize public procurement records from government agency websites, storing structured data in PostgreSQL.
-Reduced analytics query latency by 38% by rewriting PostgreSQL joins, adding targeted B-tree indexes on high-cardinality columns, and introducing a Redis caching layer for frequently accessed procurement datasets.

NeuronicJan 2026 – Mar 2026

React, FastAPI, PostgreSQL, Redis, Celery, AWS

-Deployed an AI-native learning platform on AWS with an async FastAPI backend behind an ALB, Redis-backed session and embedding caches, and a React 19 SPA on CloudFront, maintaining a sub-200ms p95 API latency.
-Built a hybrid retrieval engine combining fastembed vector search with full-text PostgreSQL indexes, powering the automatic knowledge graph construction across the full note corpus.
-Designed a multi-tenant collaboration system with RBAC, edit-suggestion workflows, real-time group feeds, integrated with Google Calendar with OAuth 2.0 and a webhook event bus.

whitematterNov 2025 – Feb 2026

C++, Python, FastAPI, Next.js, CUDA, AWS

-Built a deep learning framework in C++ with automatic differentiation, 20 layer types (Conv2D, MultiHeadAttention, LSTM) and GPU backends for Metal and CUDA.
-Implemented SIMD-optimized tensor ops (AVX/NEON) and OpenMP-parallelized GEMM hitting 99%+ MNIST accuracy in 3 epochs.
-Designed a browser-based training platform where users describe a model in natural language, agents generate the model architecture, and the system compiles and runs optimized C++ training code with real-time loss/accuracy streaming via SSE.
-Shipped as a multi-stage Docker container with a FastAPI job queue, distributed training workers, ONNX export, mixed-precision FP16, and one-click deployment to AWS EC2 GPU instances for inference saving.

Languages: Python, TypeScript, JavaScript, C/C++, SQL (Postgres)

Frameworks: React, Node.js, Flask, FastAPI

Tools: Git, Docker, Google Cloud Platform, Jira, AWS

Libraries: pandas, NumPy, Matplotlib, PyTorch