Guogang Fu

Built a production RAG pipeline (Pinecone vector store, OpenAI embeddings, LangChain) for a compliance Q&A assistant handling 400K+ monthly queries with <200ms p95 latency and full audit-trail logging for CFPB review. Designed multi-step tool-calling agent workflows (Python, LiteLLM) with memory and retry/error-handling to automate income verification and fraud-signal aggregation across IRS TDS and bank data APIs. Fine-tuned an open-source foundation model into a travel-focused conversational assistant used across multiple white-label products. Architected a SOC 2-compliant AI agent (TypeScript/NestJS) for mortgage document extraction and underwriting decision support, reducing manual processing time by 61% and time-to-close by 3 weeks. Re-architected a borrower-facing portal on Next.js 14 (React Server Components), improving Lighthouse from 51 to 93 and reducing drop-off by 38%. Implemented Go microservices with Kafka for real-time loan event streaming (80K events/sec at sub-10ms late

Built ML classification and automation to triage and categorize business platform data arriving in multiple formats, reducing operations manual effort. Developed a Tableau dashboard to visualize platform usage and user data, saving 10 hours/week in manual reporting. Trained supervised image classification for automated visual categorization and prepared POS-tagged training corpora for downstream NLP work. Built a real-time sentiment analysis model for students aged 13–16 to help teachers gauge engagement and adjust lessons. Contributed to enterprise monitoring/upgrade work using Nuxt/Vue and .NET Core with MSSQL. Delivered an MVP for lawyers to draft case reports using document extraction, analysis, classification, and OpenAI APIs (GCP, NestJS, Angular). Built and shipped web and mobile platforms including a React/Next.js property management portal (SSR/ISR, strict TypeScript) and a React Native maintenance-tickets app with push notifications and real-time tracking. Engineered backen

Developed backend services (Node.js/Express and Go) for Canvas LMS integrations, including gradebook sync, submission pipelines, and LTI integrations used by 30M+ learners globally. Built FERPA-compliant Laravel APIs for institutional billing and course licensing with field-level audit logging. Shipped React Native modules for offline content sync and biometric login with 78% adoption among active users. Implemented real-time WebSocket/MongoDB notification systems to reduce instructor response latency by 44%, and addressed GraphQL N+1 issues to reduce page load from 7.2s to 1.3s. Also built proof-of-concept AI tooling inspired by GPT Researcher using open-source components for RAG, vector search, and containerized deployment.