Engineer.
Researcher.
Builder.

M.S. Information Networking

Focus on AI systems, distributed systems, and high-performance computing.

INI Fellowship Scholarship

B.E. + B.S. Computer Science — Honours double degree

First Class Honours · GPA 3.91 / 4.0

Dean’s List, 2021–2023 · Highly Commended

AI Systems Engineering Intern / Project Lead · Covina, CA

Led a multi-tenant enterprise AI knowledge platform for a financial institution, supporting 10K+ documents and 100+ users with tenant-isolated access.

• Built hybrid retrieval with semantic chunking, batched embeddings, pgvector HNSW, keyword search, and cross-encoder reranking.
• Designed modular RAG orchestration for OpenAI and local vLLM models, including citation verification and low-confidence fallback.
• Reduced P95 query latency by approximately 40% and external API calls by approximately 50% under benchmark workloads.

Algorithm Developer · Hong Kong SAR

Built a real-time gesture-recognition pipeline spanning data collection, annotation, temporal modeling, training, and inference across approximately 20K labeled sequences and 12 gesture classes.

• Used keypoint-based CNN and LSTM/TCN models, temporal smoothing, normalization, and augmentation to improve macro-F1 from approximately 0.78 to 0.89.
• Reduced false gesture triggers by approximately 35% and reached approximately 15–20 ms per sequence on target hardware.

Lead Architect & Developer

Designed an end-to-end PyTorch architecture combining a shared causal Transformer, 24 independently parameterized recurrent agents, a learned population gate, and a probabilistic transition model. Trained on Optiver datasets with 19 microstructure features and 129-step sequences.

Built recursive stochastic rollouts and a public forward-simulation demo. Authored a technical paper and presented at the BigQuant 2026 Global Competition North America Final at UC Berkeley.

2nd place · AI Innovation track

Project Owner / Algorithm Developer

Designed a distributed market replay architecture with a metadata control plane, chunk storage, execution workers, and client interfaces. Explored deterministic replay, recovery, and separation of Go scheduling from performance-sensitive C++ execution.

Coursework assessment · 340 / 340

Project Proposer & Developer

Built a CUDA Longstaff–Schwartz engine for American options over 50K+ simulated paths. Optimized memory layouts, random-number generation, payoff kernels, reductions, and host-device synchronization.

Reported a 60× speedup over the CPU baseline with approximately 1% pricing error, supported by bootstrapping and variance-reduction experiments.

CMU parallel-computing coursework · 100 / 100

Founder & ML Engineer

Built a full-stack stateful AI product with eight interaction modes and an orchestration layer for 12 philosopher/persona agents. Designed persistent conversational history, rolling memory, parallel model calls, and independent model, persona, and storage layers.

Accepted by the Swartz Center for Entrepreneurship · 100+ pilot users