Aryan Philip
La Jolla, San Diego
Experience
Machine Learning & Application Security Intern at Shein Technology Inc.
2022-06 - 2022-08 | Palo Alto, California
- Fine-tuned B parameter Vision Language Model using PyTorch with custom loss functions, improving product filtering precision by % and reducing manual review costs by$K daily.
- Designed STRIDE threat model framework for company-wide AI security, performing SAST, DAST, and SCA analysis on LLM-based systems to identify vulnerabilities.
- Built production Flask API with SAM v for design recoloring, serving K+ daily requests with sub-s latency using Docker on AWS.
- Implemented logging infrastructure using AWS CloudWatch for attack mapping and forensic investigation, enabling automated alerting for incident response.
Software Machine Learning Intern at HCL Technologies Ltd.
2022-02 - 2022-05 | Bengaluru, India
- Developed distributed RAG system using Python/LangChain processing M+ documents across + hospital systems, achieving sub-ms response times with async processing.
- Architected vector database integration (Pinecone, Weaviate) with optimized indexing, reducing query latency by % and enabling horizontal scaling.
- Built threat detection using SHAP analysis for anomaly detection and model drift, enabling threat hunting across healthcare environments.
- Designed scalable backend on AWS with CI/CD pipelines and Docker, ensuring operational efficiency through monitoring and performance optimization.
Education
Masters of Science in Data Science in Data Science
University of California San Diego 2019-09 - 2022-06
B.Tech in Mechatronics (Robotics) in Mechatronics (Robotics)
Manipal Institute of Technology 2015-07 - 2019-04
Skills
Languages: Python, C++, Java, JavaScript, SQL, Bash, R Frameworks: MITRE ATT&CK, OWASP LLM, NIST, Security Operations Tools: PyTorch, TensorFlow, LangChain, Docker, Kubernetes, CI/CD Other: LLMs, RAG, Agents/Tools, Prompts, Embeddings, STRIDE Threat Modeling, Prompt Injection Defense, LLM Security, SAST/DAST/SCA, AWS, Azure, PostgreSQL, Redis, Vector Databases (Pinecone, Weaviate), Logging Infrastructure
Projects
Causal Copilot
Technologies: Python, PyTorch, LLMs
- Built LLM agent system with OpenAI API and tool calling, enabling natural language queries for causal relationship discovery.
- Optimized graph algorithms with C++/Python, achieving % faster computation and % fewer false positives on large-scale datasets.
- Implemented detection mechanisms for adversarial attacks and logging infrastructure, enabling forensic investigation of model behavior.
Traffic Flow Forecasting & Ramp Optimization
Technologies: Python, C++, PyTorch
- Built real-time forecasting system using Python/C++ with diffusion-convolutional GNNs, improving prediction accuracy by % for Caltrans operations.
- Designed scalable distributed system with concurrency handling and resilient architecture, ensuring reliability and operational efficiency.
- Optimized pipeline reducing latency to under % through vectorization and parallel processing, enabling real-time traffic control.