About Me

I'm Joshua Wee, a Year 4 Computer Science student at NUS with 1.5 years of backend SWE experience. I am currently interested in building at the intersection of data infrastructure and AI agents.

Patsnap

Series E Unicorn AI-powered intelligence platform used by research and development (R&D) and intellectual property (IP) teams to analyze technology trends and manage innovation.

I am currently researching and designing a flexible and modular data system to update agentic RAG pipelines.

Mitra Chem

Series B USA Battery software startup building the North American Battery Materials Champion.

I worked on data infrastructure (data unification microservices, ML featurizing pipelines, agentic copilots for data scientists).

Sparky - Data Science Agentic Copilot

A data science agentic copilot connected to data infrastructure microservices, designed to empower in-house chemical engineers and data scientists with seamless analysis capabilities.

Sparky Architecture Diagram
Click to expand

🛠️ Core Capabilities

📚 Documentation Search

Access comprehensive company scientific process documentation

💡 Code Examples

Search through common access patterns of internal microservices

🔍 AST Indexing

Abstract syntax tree indexing for enhanced code comprehension

🎯 Smart Ranking

Cross-encoder result reranking for intelligent RAG system

⚡ Code Execution

Direct database connectivity for data retrieval, EDA, and Plotly visualizations

🏗️ Technical Architecture

Framework

Built on LangGraph for orchestration and workflow management

Multi-Index System

Separate indices for user docs, examples, and code repositories

Vector Storage

FAISS implementation for efficient similarity matching

Relevance Optimization

Cross-encoder reranking with configurable weights for precision

Sample work

Projects

Singapore Hawker Centre Discovery Platform

In Development AI/ML Multi-Agent NLP

An AI-powered discovery platform that transforms how people explore Singapore's 100+ hawker centres by intelligently extracting local insider knowledge from thousands of Google Maps reviews. Processes 50K+ daily reviews to surface authentic tips like "Uncle runs out of char siu by 2pm on weekends".

Key Innovation

  • Multi-Agent AI System - Specialized agents for stall attribution, local knowledge extraction, and cultural context
  • Multilingual NLP - Processes Singlish, Chinese, Malay expressions and Singapore food terminology
  • Cultural Preservation - AllTrails-style food discovery prioritizing authentic local experiences

Tech Stack

FastAPI • PostgreSQL/PostGIS • LangChain • BrightData SERP API • Redis/Celery

View Live Project Prototype

Souris.ai

In Development AI/ML Agents

Build, schedule, and orchestrate Claude agents with a powerful desktop app. Turn your Claude Code into a production-ready AI automation platform.

Key Innovation

  • Natural Language Builder - Describe what you want in plain English. Souris generates complete agent workflows with proper tool configurations
  • MCP Integration - Auto-discover and configure MCP servers. Connect to any tool or service through a unified interface
  • Granular Permissions - Agent-level sandboxing ensures each agent only gets the tools and permissions it needs to operate

Tech Stack

TypeScript • Python

Visit Souris.ai