I am Deng Yaqi. This site records my learning and practice across NLP, knowledge graphs, Java backend engineering, and AI engineering, including RAG, document parsing, agents, retrieval quality, and practical system design.
Hangzhou Genaima Network Technology Co., Ltd. / Hangzhou, China
Leads AI R&D planning, technical direction, and project delivery, bringing enterprise AI platforms, Agent/RAG workflows, data governance, and business copilots from zero to one.
iSoftStone Technology Service Shanghai Co., Ltd. / Shanghai, China
Developed RAG systems, vector knowledge bases, intelligent Q&A engines, and AI coding platforms, moving AI tools from proofs of concept toward deliverable applications.
Beijing Hanke Times Technology Co., Ltd. / Ant Group MYbank Risk Control
Worked onsite with Ant Group's MYbank risk-control team, developing backend capabilities for strategy R&D, industry risk management, and risk strategy platforms.
Zhejiang Tsinghua Flexible Electronics Technology Institute / Hangzhou, China
Built medical Android software end to end, covering design, coding, debugging, device interaction, cloud communication, and post-release support.
Built an enterprise AI backend with FastAPI, LangChain, LangGraph, OpenAI SDK, PostgreSQL, ClickHouse, Neo4j, and Redis for multi-model access, RAG knowledge bases, MCP tools, and business integration.
Built customer research, search, report generation, review, sharing, and follow-up workflows with LangGraph and business prompts, plus Agent/LLM/Tool audit traces.
Built a Windows/macOS activity collector that writes local behavior events to SQLite/JSONL and PostgreSQL for dashboards, daily reports, and work-process analysis.
Built customer, demand, and product cards for cross-border ecommerce operations, connecting customer profiles, demand recognition, and product recommendations.
Built a Vue 3, Vite, and Chrome Extension MV3 sidebar that embeds AI platform capabilities into business pages such as WhatsApp, with customer context, message history, draft replies, and response suggestions.
Built a workflow for ICF installation PDFs, including PDF-to-image conversion, AI information extraction, OSS upload, Excel comparison, and API queries.
Why prompt injection is dangerous for agents, and how to build defenses with untrusted content isolation, least privi...
An engineering view of Agent Harness Engine and why agents are runtime systems built from context, tools, permissions...
A first-time developer guide covering download, installation, first launch checks, and safety notes.
A first-time guide to installing Node.js, searching skills, downloading them into a Codex project, and validating ins...
A system-by-system guide for first-time Codex users covering install entry points, commands, verification, and common...
Notes on document parsing, structure preservation, chunk validation, and open-source reproduction before RAG ingestion.