Shane (Xingeng) Zhang

Co-Founder & Lead Engineer | AI/ML Specialist

Summary

I design and build end-to-end AI agent infrastructure that turns labor-intensive workflows into reliable, scalable systems.

Work Experience

Co-Founder & Lead Engineer

Jan 2025 - Present

Empath Legal

Piscataway, NJ

Building AI-powered legal research tools with my co-founder Grant, focused on creating systems that accelerate lawyers' analysis without replacing their judgment. Bridging the gap between complex AI capabilities and non-technical professionals who need user-friendly, reliable tools for high-stakes decision-making.

  • Architected and implemented an advanced Agentic RAG system with Pydantic AI, orchestrating multiple GenAI agents using the ReACT framework for improved content retrieval accuracy in legal document analysis.

  • Designed and developed a vertically scalable RESTful API backend with Litestar (modern Python framework with superior architectural flexibility compared to FastAPI), utilizing WebSocket and Server-Sent Events for async-reactive real-time LLM integration.

  • Implemented comprehensive evaluation frameworks to ensure high-quality AI-generated outcomes, reducing hallucination rates and maintaining reliability for legal professionals.

  • Integrated multiple LLM providers including Google Gemini (for long-context processing), Anthropic Claude (for output quality), OpenAI GPT (for personality and reasoning), and open-source models like Mixtral for cost-efficient solutions.

  • Collaborated on frontend development using TypeScript, SvelteKit, and shadcn to create highly intuitive, user-friendly interfaces designed specifically for non-technical lawyers, emphasizing ease of use and interactive workflows.

  • Designed and maintained PostgreSQL database architecture with proper indexing strategies, utilizing SQLAlchemy and Advanced Alchemy ORMs to optimize query performance for high-volume AI operations.

  • Built proprietary GenAI framework for internal use to ensure high-quality, predictable outputs across different LLM providers and use cases.

  • Fully containerized the solution using Docker for scalable deployment, implementing comprehensive CI/CD pipelines with GitHub Actions, SonarQube, and PyTest for AsyncIO testing.

  • Engineered workflow orchestration using Celery for task management and Temporal Workflow for handling complex, reliable multi-step processes.

  • Implemented Redis for PubSub capabilities and caching, and integrated OpenSearch for observability (metrics and traces) alongside structlog for comprehensive logging.

  • Designed sophisticated prompts and guardrails to ensure AI assists users without nudging or misleading their opinions, keeping human judgment central to decision-making processes.

  • Led cross-functional team meetings with non-technical stakeholders to ensure AI development aligns with customer needs and delivers meaningful value to legal professionals.

  • Emphasized strongly typed Python throughout the codebase to ensure robustness, maintainability, and reliability in production environments.

AI/ML Engineer

Aug 2024 - Dec 2024

Citigroup Inc.

Rutherford, NJ

Developed enterprise-scale RAG systems for financial compliance and document analysis, working with structured data processing pipelines and multi-cloud deployments.

  • Built structured data processing pipelines with Python for efficient document handling and analysis.

  • Designed RESTful APIs with FastAPI for real-time data ingestion and NLP processing into PostgreSQL databases.

  • Deployed solutions at scale with OpenShift and Apache Spark, maintaining robust CI/CD procedures.

  • Leveraged Agentic RAG systems and Knowledge Graphs (RDF and LPG formats) using LangChain to improve retrieval accuracy in AI-driven compliance analysis.

  • Implemented evaluation metrics and visualizations to ensure LLM quality and reduce error investigation time.

  • Utilized OpenAI GPT, Claude, and Google Gemini models, and fine-tuned open-source LLMs to develop customized Agentic RAG systems.

  • Developed human-centered evaluation frameworks (RLHF) to assess LLM performance in real-world scenarios and ensure alignment with user intents.

  • Deployed AI-driven RAG systems on AWS using S3, EC2, Glue, Lambda, SageMaker, and Bedrock for data processing and LLM integration.

  • Developed FastAPI interfaces and gRPC protocols for API integration across Azure and AWS cloud providers.

AI/ML Engineer

Jan 2023 - Aug 2024

Robert Wood Johnson University Hospital

New Brunswick, NJ

Built AI-powered research assistant tools for medical researchers, automating document analysis and improving research workflows for academic teams.

  • Designed, built, and deployed an Agentic RAG system using Python, JavaScript, SQL, and Chroma vector database to automate parsing and summarization of research documents.

  • Engineered an advanced RAG system with OpenAI and React framework to reduce onboarding time and improve research efficiency for academic staff.

  • Collaborated with a cross-functional team of academic researchers, integrating feedback into the NLP system to align the platform with research objectives.

  • Conducted workshops on ML usage for staff members, driving adoption of ML-enhanced workflows across the organization.

  • Utilized AWS SageMaker for model fine-tuning and AWS Bedrock for serving models in production environments.

Machine Learning Engineer

Jan 2020 - July 2021

Fiskkit Inc.

San Francisco, CA

Integrated NLP-driven features into a production web platform, focusing on real-time text generation and semantic analysis using PyTorch and graph databases.

  • Integrated NLP-driven features into the Node.js backend with PyTorch (C++ CUDA) for real-time text generation and summarization.

  • Conducted data pre-processing and exploration using PySpark, NumPy, and Pandas to ensure high-quality data integration for model training.

  • Optimized deep learning models through quantization and pruning techniques with TensorRT.

  • Built a semantic graph database using Neo4j and Cypher queries to store and query complex relationships.

Teaching

  1. Teaching Assistant
    Rutgers University
    Dec 2020 - Dec 2022

    Tutored students in Computer Science and Machine Learning, helping bridge theoretical concepts with practical implementation

Publications

Publications

  • Berns, M. P., Nunez, G. M., Zhang, X., et al. (Sep 2024). Auditory Decision-making Deficits After Permanent Noise-induced Hearing Loss.