Julian Wong

Led AI-assisted revenue operations workflows for B2B SaaS customers using OpenAI API and HuggingFace models, reducing manual CRM updates by 38%. Designed industry-tailored RAG architectures and LLM orchestration for enterprise sales docs, improving domain-specific answer relevance by 29%. Orchestrated multi-step LLM workflows across proposal drafting and customer follow-up processes. Implemented prompt evaluation and observability using LangSmith and MLflow, reducing hallucination incidents by 26% through iterative testing. Built embedding and inference pipelines with Python, Pandas, and Kafka, processing 500K+ customer interactions for semantic indexing. Experimented with lightweight fine-tuning using PyTorch and HuggingFace Transformers to boost domain accuracy by 17%. Deployed scalable LLM inference endpoints using FastAPI, Docker, Kubernetes, and vLLM, maintaining sub-250 ms response latency under concurrent user load.

Modeled enterprise authorization graphs using Neo4j, Python, and Java, processing 50M+ access relationships to support identity visibility workflows. Orchestrated streaming and batch ingestion pipelines with Kafka, Spark, and Airflow, increasing data synchronization throughput by 41% across multi-tenant environments. Established data normalization layers with Pandas and Python, improving schema validation accuracy for enterprise connectors. Built backend services with Python (Django, FastAPI) and Ruby on Rails, accelerating feature delivery by 27% through modular service boundaries. Implemented role-similarity scoring prototypes with Scikit-Learn for security teams, reducing scanning time on 10K+ identities. Provisioned containerized workloads with Docker and Kubernetes, reducing staging release rollbacks by 23%. Introduced semantically-driven search experiments using HuggingFace embeddings and vector indexing to cut investigation query time by 35%. Consolidated monitoring and audit pi

Delivered high-performance trading interfaces using React.js and TypeScript, supporting 1M+ daily active users during peak market sessions. Integrated real-time market feeds through WebSocket streams and client-side state handling, decreasing UI re-render latency by 37%. Configured CI/CD workflows with GitHub Actions and Docker-based builds, shortening frontend deployment cycles from 45 minutes to 18 minutes. Enhanced data-fetch architecture with React Query, eliminating 42% of redundant API requests across portfolio and order views. Strengthened production observability using DataDog and Sentry, reducing mean time to detect frontend incidents by 33%. Coordinated API schema updates with backend services through contract validation and feature flag rollouts, lowering production regression rates by 24%.