💼 I’m currently a Software Engineering Intern @ Google

• Interning on the GCP team in the Greater Seattle Area in Summer 2025.

🔭 Previously, I was a Software Engineer Intern @ Zilliz

• Co-developed DeepSearcher, a Retrieval-Augmented Generation (RAG) framework with 6k+ GitHub stars, designed for private data QA using LLMs and the Milvus vector database.
• Integrated Milvus Vector Database into 20+ open-source projects and authored docs & tutorials to drive developer adoption and community engagement.
• Enhanced existing Milvus integrations by incorporating BM25-based full-text search for hybrid lexical-semantic retrieval and migrating from the legacy ORM SDK to the new MilvusClient SDK.

🧠 Earlier Work Experiences:

• Machine Learning Engineer Intern @ Pinduoduo

  • Enhanced mContriever models for AI-powered shopping assistants, achieving a +27% recall and +21% F1 score.
  • Developed embedding pipelines and optimized RAG generation, reducing latency by 7.6%.

• Software Engineer Intern @ Capybara.AI

  • Built a concurrent system for real-time news collection, analysis, and sentiment scoring.
  • Designed industry-specific clustering algorithms and engineered tools for financial analysis, integrating sentiment analysis into sector-wide predictions.

• Multimodal Machine Learning Engineer Intern @ Vipshop

  • Fine-tuned VisualGLM’s Q-Former with advanced attention masking techniques, reducing attribute mismatches by 6.79%.

Interview AI: Resume-aware mock interview platform with OpenAI TTS/STT, ELO scoring, AI-suggested answers, and flashcard review.

FireFrontAI: Deep learning model for next-day wildfire spread using Presto + U-Net; outperformed UTAE/ConvLSTM on mNDWS with Jaccard loss and fire priors.

SparTMoE: Sparse attention-gated MoE model using Switch Transformer and sparsemax routing; improved ROUGE on SAMSUM with minimal overhead.

LDCT Denoising: Benchmarked UNet variants for low-dose CT denoising.

Ecowise: Gamified sustainability platform with appliance detection via Cloud Vision, energy usage insights, OpenAI-powered chatbot, and quiz-based engagement.