Wei Chen
APPLIED AI / ENGINEERING PORTFOLIO
identity.tsUTF-8

portfolio@local:~$ cat profile

Wei Chen

Applied AI
Engineer.

Product-focused Software Engineer building scalable, reliable AI apps with Python, Node.js, LLMs, and AI agents, using prompt engineering, tool calling, and APIs to automate tasks and solve real customer problems.

LLM Applications RAG Systems AI Agents
Explore projects View Resume
EmailGitHubLinkedInPortfolio
HOW I APPROACH AI

Documents in.
Actions out.

The engineering between a document and a useful action: shaping context, choosing tools, and checking what comes back.

  1. 01

    Shape the evidence

    Ground answers in relevant documents with retrieval and hybrid search.

  2. 02

    Connect context to tools

    Connect language models, tools, and APIs to solve the task.

  3. 03

    Check the result

    Validate structured outputs and handle failures before returning a result to the user.

Three floating circuit layers connect document fragments to a context graph and tool outputs.
THE INFERENCE STACK / FROM DOCUMENT TO ACTIONDrag to rotate. Arrow keys change the view. Home restores the starting view.
01 / CAPABILITIES

From model to product.

Connecting AI capabilities to the software people use.

01

Reason

LLM applications, tool calling, and agent workflows.

Python
LLM APIs
AI Agents
02

Retrieve

Relevant knowledge through retrieval and hybrid search.

RAG
Hybrid Search
PostgreSQL
03

Build

Reliable APIs and interfaces around real customer needs.

TypeScript
Node.js
Next.js
04

Ship

Cloud infrastructure and automated delivery pipelines.

AWS
Docker
GitHub Actions
02 / EXPERIENCE

Engineering in practice.

Myanmar First Financial Group Co., Ltd.Senior Software EngineerYangon, MyanmarOctober 2025 – Present

Financial services group · Engineering team

  • Develop and maintain a React, Next.js, and Node.js customer dashboard serving 1,000+ business accounts.
  • Built an admin panel with a secure enterprise RAG assistant using hybrid search and role-based access control: 85% answer accuracy and 90% less document search time.
  • Mentored two junior engineers through code reviews and pair programming.
React
Next.js
Node.js
RAG
Hybrid Search
RBAC
Shenzhen Tengdaxun Technology Co., Ltd.Software EngineerShenzhen, ChinaNovember 2020 – July 2025

B2B access-control solutions · Five-person engineering team

  • Developed REST APIs with Node.js, Express, and MongoDB for orders, billing, invoicing, and customer support.
  • Migrated 40+ API endpoints from AWS EC2 to Lambda, reducing hosting costs by 30% while supporting 100 concurrent requests.
  • Built a customer-facing RAG chatbot with hybrid search: 90% answer accuracy and 95% faster responses compared with manual support.
  • Built an OCR and DeepSeek purchase order workflow: 80% extraction accuracy and 70% less manual data entry time.
  • Built AWS CI/CD pipelines with GitHub Actions and Docker, reducing deployment time from 40 to 6 minutes.
Node.js
Express
MongoDB
AWS Lambda
DeepSeek
Docker
Ningbo Yilian Huishang Information Technology Co., Ltd.Junior Software EngineerNingbo, ChinaSeptember 2018 – October 2020

Web development

  • Developed responsive websites for 10+ clients using React and Node.js.
React
Node.js
03 / PROJECTS

Projects

Built, shipped, and ready to explore.

KnowledgeBase website showing its knowledge assistant alongside a cited source document.Explore KnowledgeBase
LIVE WEBSITE

KnowledgeBase

Your documents. Answers with evidence.

A RAG workspace for bringing team knowledge together, asking questions, and following citations back to the original documents.

  • Organized knowledgeDocument collections keep conversations focused.
  • Hybrid retrievalSemantic and keyword search with reranking.
  • Traceable answersPage-level PDF citations connect answers to sources.
Visit website (opens in a new tab)
04 / SELECTED WORK

Project case files.

Real applications. Measurable outcomes. A closer look at the systems behind the work.

Showing 4 projects

CASE_01RAG systems
CONCEPTUAL WORKFLOW01
Employee query
Access control
Hybrid retrieval
Grounded answer
Enterprise AI / RAG

Enterprise RAG Knowledge Assistant

Internal knowledge, made accessible. A secure assistant helping financial services employees find answers in company documents.

BUILT WITH
RAG
Hybrid Search
RBAC
Document Retrieval
85%answer accuracy
90%less search time
Read case study
CASE_02RAG systems
CONCEPTUAL WORKFLOW02
Customer question
Approved documents
Hybrid retrieval
Support answer
Applied AI / Customer Automation

AI Customer Support Chatbot

Product answers without the wait. A customer-facing assistant grounded in approved product and support documentation.

BUILT WITH
RAG
Embeddings
Hybrid Search
LLM Integration
90%answer accuracy
95%faster responses
Read case study
CASE_03Automation
CONCEPTUAL WORKFLOW03
Purchase order
OCR
DeepSeek extraction
Structured data
Document AI / LLM Automation

AI Document Processing Pipeline

From scanned pdfs and images to structured information. An OCR and DeepSeek workflow that reduces repetitive data entry.

BUILT WITH
OCR
DeepSeek
Structured Extraction
API Integration
80%extraction accuracy
70%less manual entry
Read case study
CASE_04AI agents
AGENT HARNESS / EXECUTION FLOW04
Scout code
Edit & probe
Review evidence
Return patch
AI Agents / Developer Tools

Database Engineering Agent & Execution Harness

A Python agent harness that turns database engineering tasks into Git patches through repository scouting, tool-driven edits, SQL probes, and evidence-based review.

BUILT WITH
Python
LLM Tool Calling
PostgreSQL
ClickHouse
Git Patch Validation
Context Management
Agent Automation
36%reported benchmark solve rate
Read case study
How these AI systems fit together
FROM REQUEST TO RESPONSE 01 — 05
User requestA question. A task. A starting point.
RetrievalRelevant context
KnowledgeApproved sources
AI agentReason · Retrieve · Act
Tool callingAPIs & functions
Useful response

Conceptual workflow · Built around real problems

Reported project outcomes. Evaluation context is noted in each case study.

05 / FIELD NOTES

Blogs

Notes on building reliable AI systems and the engineering decisions behind them.

A computer monitor displaying analytics charts
RAGEvaluation

Evaluate RAG at the Retrieval Boundary

Separate missing evidence from unsupported answers using relevance labels, retrieval metrics, and a small, reviewable evaluation set.

· ~2 min read
Bookshelves and hanging lights in a library
RetrievalSearch Engineering

Hybrid Search: Give Exact Matches and Meaning a Vote

Why lexical retrieval still matters, how reciprocal rank fusion combines results, and what to measure before adding a reranker.

· ~2 min read
A close-up photograph of a computer circuit board
LLMsContext Engineering

A Larger Context Window Is Not an Evidence Strategy

What long-context research tells us about evidence placement, and how to budget context without turning every request into a document dump.

· ~2 min read
Eyeglasses in front of computer screens displaying source code
AI AgentsBackend Engineering

Tool-Calling Agents Need Explicit Execution Boundaries

Design tools as contracts: validate arguments, enforce authorization, bound retries, and inspect outcomes outside the language model.

· ~2 min read
Visit the writing archive
06 / RESUMES

Resumes

A focused view of my experience, tailored to the work your team needs.

07 / CONTACT

Let’s build something useful.

Have an AI workflow, product idea, or engineering challenge in mind? Let’s start a conversation.

Let’s talk about your project

Share the challenge you’re working on and what a useful result would look like.

PDF, JPG, PNG, DOCX, or TXT · Maximum 3 MB

Your details and attachment will be used to respond to this inquiry. Privacy Policy