AI Engineer focused on building production-grade AI systems including LLM agents, Retrieval-Augmented Generation (RAG) pipelines, and computer vision applications.
🎓 Master of AI graduate (Australia Awards Scholar) specializing in Natural Language Processing
🔬 Currently building AI systems combining LLMs, search, and vision models
📍 Targeting Human-Centered AI and ML Engineering roles in Singapore, UK, and Remote
Routes plain English questions to SQL, industry documents, or both — automatically.
How it works:
- 🗄️ SQL route — "What is our total revenue by genre?" → queries Chinook DB → Rock $826.65
- 📄 Document route — "What is global music revenue growth?" → searches Pinecone → +4.8% (IFPI 2025)
- 🔀 Both route — "How does our Rock revenue compare to global trends?" → combines both sources
Technical highlights:
- LangGraph state machine (7 nodes, 3 routing paths)
- Pinecone vector DB — 2,462 vectors from 4 real industry PDFs
- LLM-as-judge evaluation framework (route-specific scoring)
- HCD features: explainability panel + human feedback loop (mirrors RLHF)
- 86.1% benchmark accuracy (36 queries, 6 tiers) · 0% SQL hallucination
Stack: LangGraph · GPT-4o-mini · Pinecone · FastAPI · Next.js · Google Cloud Run
🔗 Repository · Live Demo
Production-grade Retrieval-Augmented Generation system for querying document collections.
Features:
- Hybrid retrieval (vector + keyword search)
- Query expansion (HyDE)
- Hallucination detection
- 74% accuracy across 38 evaluation queries
Unified backend for multiple AI providers powering 5 AI products.
Capabilities:
- Single API for multiple LLM providers
- Automatic failover (Groq → Gemini → OpenAI)
- Real-time cost tracking
- Session management (in-memory + Redis)
Supports: OpenAI | Gemini | Groq
Computer vision system for real-time object detection and tracking.
Tech:
- Mask R-CNN for detection
- SORT tracking algorithm with Kalman filtering
- Self-supervised evaluation metrics
- 78.4% tracking accuracy with 100% ID stability
| Area | Skills |
|---|---|
| LLM Applications | Agentic RAG, LangGraph, LLM-as-judge Evaluation, Human-Centered AI, Tool Use, Function Calling |
| Search & Retrieval | Hybrid Search (Vector + BM25), Pinecone, Query Expansion, ChromaDB |
| Prompt Engineering | Few-shot, Chain-of-Thought, System Prompts |
| Computer Vision | Object Detection, Tracking, Medical Imaging, Foundation Models (SAM) |
| Infrastructure | API Orchestration, Multi-provider Failover, Session Management |
| Category | Technologies |
|---|---|
| Languages | Python |
| AI/ML | PyTorch, TensorFlow, Transformers, LLMs, NLP, Computer Vision |
| LLM Ecosystem | OpenAI, Anthropic, Groq, Google Gemini, LangChain, LangGraph |
| Backend | FastAPI, REST APIs, Redis |
| Tools | Docker, Git, CI/CD (GitHub Actions), VS Code, Jupyter |
