Hi, my name is
Barbara Wu.
|
AI engineer with 7+ merged PRs into LangChain, Haystack, and CAMEL. I turn LLMs into production systems — RAG pipelines, intelligent agents, AI-powered workflows — on frameworks I know from the inside out. Trilingual (Chinese, French, English) with a PhD from Sorbonne University. Trained to think in systems, driven by curiosity to understand why — whether it's an LLM's reasoning or my own.
Ask My AI
Powered by LLM · trained on my resume
I'm an AI trained on my resume, projects, and experience. Ask me anything!
01
Projects
Full-Stack RAG Conversational AI
RAGTalk
Real-time conversations with historical figures grounded in primary sources. Agentic query routing, hybrid BM25+vector search, Reciprocal Rank Fusion, cross-encoder reranking, and SSE streaming.
Multi-Agent Research Assistant
AI Researcher
AI-powered research assistant with LangGraph orchestration. Self-Refine, Reflexion, Corrective RAG, and Multi-Agent Debate patterns for deep research synthesis.
Enterprise Support Platform
IT Ticketing System
Internal IT support platform with AI-powered ticket analysis, async task processing, 4-language internationalization (CN/EN/FR/NL), and full observability stack.
Other Noteworthy Projects
Personal Portfolio
This site! Built with Next.js 16, Matter.js physics engine, AI chatbot, and Tailwind CSS.
Open Source Contributions
7+ merged PRs to LangChain (100k★), Haystack (18k★), CAMEL (10k★). Streaming fixes, memory leaks, API design.
02
Experience
Contributing Developer @ Open Source
Feb 2026
- Merged PRs into LangChain (100k★), Haystack (18k★), CAMEL (10k★) production codebases
- Fixed streaming bugs, memory leaks, and data loss issues in major LLM frameworks
- Designed API features including a flexible document comparison field for RAG evaluators
- Wrote 7 in-depth technical blog posts analyzing each contribution's root cause and fix
- Iterated on maintainer feedback to ensure backward compatibility and comprehensive test coverage
Technical Skills
Languages
Frontend
Backend
AI / ML
Infrastructure
03
Blog
Building an Agentic RAG System from Scratch: Architecture, Hybrid Search, and Adaptive Query Routing
A deep dive into RAG Talk — a conversational AI system with hybrid BM25+vector search, Reciprocal Rank Fusion, cross-encoder reranking, and an intelligent agentic routing layer that adapts retrieval strategy to query intent.
Building a Production-Grade IT Ticketing System: Django, React, Celery, and AI-Powered Ticket Analysis
A comprehensive technical walkthrough of designing and building an enterprise IT support platform with async task processing, 4-language i18n, RAG-powered ticket analysis, and a full observability stack with Prometheus and Grafana.
How I Fixed a Streaming Bug That Corrupted Parallel Tool Calls in LangChain
A deep dive into LangChain's merge_lists() function, why parallel tool calls from Bedrock/Anthropic got corrupted, and designing a backward-compatible fix at the framework level.
Hunting Down a Memory Leak in LangChain's lru_cache
How Python's descriptor protocol creates new method objects on every access, causing @lru_cache to have 0% hit rate but 100% memory retention. The fix: 15 lines.
Adding Structured Reasoning to Haystack's Streaming Pipeline
Migrating Anthropic and Ollama integrations from unstructured meta dict access to proper ReasoningContent objects, with helper extraction and accumulation refactoring.
Designing a Flexible Document Comparison API for Haystack Evaluators
Adding a document_comparison_field parameter to MRR, MAP, and Recall evaluators — enabling comparison by ID or metadata instead of hardcoded content matching.
Extracting Hyperlinks from PPTX and XLSX Files in Haystack
How pandas.read_excel() silently discards hyperlinks, and the openpyxl workaround to preserve them. Plus python-pptx run-level hyperlink extraction.
Fixing Lost Rich Output in CAMEL's E2B Code Interpreter
An early-return chain silently dropped matplotlib plots and images. The fix: replace it with an accumulation pattern that collects output from all sources.
Improving Error Messages for Non-OpenAI Embedding Providers in LangChain
When OpenAIEmbeddings is used with third-party providers, cryptic errors replace helpful guidance. The fix: better error messages and documentation.
04
Writing
On style, imitation, and what machines know about literature.
The Pasticheur’s Machine
What nineteenth-century literary theory knows about large language models, and what LLMs reveal about style, originality, and imitation.
Pastiche: After the Manner of Vladimir Jankélévitch
A long-form essay that performs what it describes: an investigation of stylistic imitation written as a pastiche of Jankélévitch’s philosophical prose.
04
Get In Touch
I'm actively looking for Developer Relations and AI engineering roles. Whether you have a role that fits, a project idea, or just want to connect — my inbox is always open.
Say Hello