✨ cv
an Aspiring AI engineer 🔥
Basics
| Name | Amy (Chujun) Tao |
| Label | Software Engineer |
| chujunt@andrew.cmu.edu | |
| Phone | 919-813-8187 |
| Url | https://AmyTao.github.io |
| Summary | Software Engineer with expertise in distributed systems, machine learning infrastructure, and full-stack development. Experienced in building scalable data pipelines, recommendation systems, and cloud-native applications. |
Work
-
2025.05 - 2025.08 Software Engineer Intern
Meta
Unified revenue-sharing workflow orchestration engines, reducing system complexity and cutting engineering onboarding time by 50%.
- Migrated and refactored 7 tasks of the digital balance spend workflow and created unit tests using gMock
- Enhanced workflow extensibility by integrating it with the Action Config Factory, enabling reuse across five types of Facebook digital products
- Implemented asynchronous DAG-based workflow execution with step-level tracing, providing end-to-end visibility through error logging and ensuring idempotent execution
- Dogfooded the fan-funding subscription purchase flow using the new workflow, validating end-to-end transactions on Google Pay and Apple Pay
Education
-
2024.08 - 2026.05 Pittsburgh, USA
-
2023.01 - 2023.08 Berkeley, USA
-
2020.09 - 2024.07 Shenzhen, China
Certificates
| NVIDIA Certificate | ||
| NVIDIA | 2024-12-01 |
Skills
| Programming Languages | |
| C/C++ | |
| PHP | |
| Java | |
| Go | |
| Python | |
| JavaScript | |
| TypeScript | |
| Bash | |
| SQL | |
| Verilog |
| Frameworks & Tools | |
| React.js | |
| Node.js | |
| gMock | |
| GraphQL | |
| Redis | |
| Kafka | |
| RabbitMQ | |
| Docker | |
| Kubernetes | |
| AWS | |
| Azure | |
| Nginx | |
| JUnit | |
| Postman | |
| Maven | |
| Git | |
| Jenkins | |
| Prometheus | |
| Grafana |
Languages
| English | |
| Fluent |
| Chinese | |
| Native speaker |
Interests
| Machine Learning | |
| ML Infrastructure | |
| AI Infrastructure | |
| LLM Application | |
| LLM Infrastructure | |
| Recommendation Systems | |
| Deep Learning | |
| Natural Language Processing |
| Distributed Systems | |
| Cloud Computing | |
| Microservices | |
| System Design |
Projects
- 2025.09 - 2025.12
CourPilot: A Multi-Source Deep Research Assistant for Course Materials
- Orchestrated a multi agent system using Claude Sonnet 4.5 for intelligent tool selection and decision-making, with semantic routing to optimize tool chains across multiple MCP servers.
- Built BERTopic-based topic modeling system integrated with Milvus vector database for semantic clustering, automatic topic reduction, and dynamic knowledge base updates with hot-reloading capabilities.
- Implemented a hybrid search architecture with fine-tuned Mistral-7B, web search (Tavily API), and academic paper retrieval, enabling hallucination detection on LLM responses via diversion decoding to measure model resistance to semantic deviation.
- 2025.09 - 2025.12
Sparse Autodiff Framework: Efficient Sparse Matrix Operations for Graph Neural Networks
- Extended the Needle Autodiff framework with sparse matrix support, implementing COO and BSR storage formats with format conversion capabilities.
- Designed and implemented a sparse-dense matrix multiplication operator (SpMM) integrated with the autograd computation graph, enabling gradient computation through sparse operations.
- Built an end-to-end GCN recommendation model on MovieLens 100K, constructing bipartite graph adjacency matrices with 97% sparsity and achieved order-of-magnitude improvements in memory efficiency (26 to 3.9MB) and computational efficiency (0.22 to 0.006 GFLOPs) by processing only non-zero entries.
- 2025.04 - 2025.05
Movie Recommender for Streaming Platform
- Designed and deployed a modular movie recommender system for 10K users and 27K movies, featuring real-time data ingestion with Kafka, automated model updates via Docker, cold-start handling, popularity-aware re-ranking, model versioning and lineage with MLflow, and observability using Prometheus and Grafana.
- Applied popularity-aware re-ranking to promote mid-/long-tail content for exposure diversity and conducted post-hoc fairness audits to mitigate bias
- Enabled traceable and reproducible deployments by leveraging MLflow for model registry and pipeline versioning
- Enhanced zero-downtime model retraining by triggering on both user behavior drift and scheduled intervals
- 2025.10 - 2025.12
Distributed Data Mining and Search Engine for Scholarly Abstracts
- Built a containerized web application that allowed users to submit Google Scholar URLs filtered for IEEE Xplore, parsed search results, and analyzed paper abstracts to provide interactive search and the top N frequent terms.
- Provisioned a scalable, cluster-based service on Google Kubernetes Engine (GKE) using Terraform
- Integrated with a user-facing application via Kafka for message streaming
- Constructed the inverted index and the top N algorithm using Hadoop MapReduce on the cluster with a stop-word list
- 2025.09 - 2025.12
Needle
- Implemented a PyTorch-like deep learning framework supporting automatic differentiation, tensor operator backends, and modular training workflows.
- Built a flexible optimizer and initialization system enabling configurable training setups
- Developed core neural network modules, including Batch Normalization, Layer Normalization, and Dropout for improved model training stability
- 2025.04 - 2025.05
Dynamic Memory Allocator (Malloc)
- Designed a highly effective dynamic memory allocator with the best-fit allocation algorithm and LIFO eviction, achieving high throughput (9574 kops/sec) and high memory utilization (73.8%).
- Improved memory utilization with segregated free lists and mini blocks to reduce internal fragmentation
- Used a coalescing strategy to reduce external fragmentation