MSc Signal Processing & Machine Learning · Nanyang Technological University, Singapore Research Attachment @ I²R, ASTAR*
Available for full-time AI / ML roles in Singapore from July 2026
I build end-to-end AI systems — from agentic pipelines and RAG APIs to LoRA fine-tuned LLMs, anomaly detection, and full-stack fintech dashboards — all production-ready with FastAPI, Docker, Kubernetes, and GitHub Actions CI/CD.
- 🔬 Research Attachment @ I²R, A*STAR — built a physics-accurate 3D ISAC digital twin (NVIDIA Sionna ray tracing) for 5G vehicle sensing; designed OFDM radar with static clutter cancellation, Doppler trajectory prediction at 0.333s intervals, and sensing-aided beamforming
- 🤖 Agentic AI — autonomous news pipeline (Gemini 2.0 Flash + ChromaDB RAG + Kubernetes CronJob); RAG document Q&A (FAISS + Groq Llama 3.3-70B + RAGAS evaluation)
- ⚡ LLM Fine-Tuning — DistilBERT + LoRA/PEFT on FinancialPhraseBank; 98.7% parameter reduction; F1 = 0.846
- 📈 MLOps — LSTM Autoencoder anomaly detection with LP scheduling (SciPy linprog), SimPy DES, KS-test drift monitor, 150-test CI/CD pipeline
- 💳 Full-Stack AI — fintech risk dashboard: 4 Spring Boot microservices + React + LSTM risk scoring + Groq LLM explainability
- 🌐 NLP — multilingual transformer fine-tuning (XLM-RoBERTa, mBERT, ALBERT); Macro-F1 0.84; 3rd place globally at DravidianLangTech@EACL 2024
Languages
ML / Deep Learning
NLP & LLMs
-F55036?style=flat){kind=icon}
Agentic AI & RAG
MLOps & Deployment
-FF9900?style=flat&logo=amazonaws&logoColor=white){kind=icon}
Optimisation & Simulation
-8CAAE6?style=flat&logo=scipy&logoColor=white){kind=icon}
| Project | Description | Stack | Key Result |
|---|---|---|---|
| 💳 Fintech Payment Risk Dashboard | 4 Spring Boot microservices + React dashboard; LSTM Autoencoder risk scoring (rule 40% + ML 60%); Groq LLM for plain-English explainability and NL→filter query parsing | Java Spring Boot, React, MongoDB, Python FastAPI, Docker, GitHub Actions | 4 microservices · hybrid ML+rule scoring · LLM NL query parsing |
| 🤖 Agentic AI Tech News Pipeline | Autonomous daily pipeline: ArXiv, HackerNews, TechCrunch, VentureBeat → relevance filter → Gemini 2.0 Flash digest → Gmail. Built-in RAG query interface over stored articles | Gemini 2.0 Flash, ChromaDB, Sentence Transformers, Docker, Kubernetes, GitHub Actions | 97 articles/day · semantic drift detection · K8s CronJob CI/CD |
| 📄 RAG Document Q&A API | PDF → section-aware chunking → FAISS → Groq Llama 3.3-70B grounded answers; hallucination risk scoring; PDFs + index on AWS S3 | LangChain, FAISS, HuggingFace, FastAPI, Docker, MLflow, AWS | 64% keyword recall · 0.63s latency · Faithfulness: 0.783 · Relevancy: 0.932 |
| ⚡ LLM Fine-Tuning with LoRA | DistilBERT on FinancialPhraseBank, 3-class sentiment via LoRA/PEFT; automated model registry with weekly retraining pipeline | HuggingFace PEFT, LoRA, FastAPI, Docker, MLflow, AWS | F1 = 0.846 · 98.7% param reduction (887K vs 67M) |
| 📈 Time-Series Anomaly Detection | LSTM Autoencoder on NASA SMAP (54 channels); LP fractional knapsack inspection scheduler; SimPy DES; KS-test drift monitor; 150-test CI/CD suite | PyTorch, SciPy linprog, SimPy, FastAPI, Docker, MLflow | F1 = 0.737 · AUC-ROC = 0.857 · LP +17.6% signal · 24% makespan ↓ |
| 🔬 Industrial Defect Detection | End-to-end defect detection on MVTec dataset; Otsu thresholding + regionprops characterisation; ResNet18 binary classification | PyTorch, ResNet18, scikit-image, OpenCV, MLflow, FastAPI | >90% accuracy · per-region defect measurements |
| 🌐 Fake News Detection | Multilingual misinformation detection in Dravidian languages; XLM-RoBERTa, mBERT, ALBERT fine-tuning | XLM-RoBERTa, mBERT, ALBERT, HuggingFace | Macro-F1: 0.84 · 3rd place globally · EACL 2024 |
| 📡 5G ISAC Digital Twin | Physics-accurate 3D urban digital twin (Fusionopolis, SG); OFDM radar + SCC at 59 GHz; Doppler trajectory prediction at 0.333s; sensing-aided beamforming | NVIDIA Sionna, Blender, OpenStreetMap, Python | First-authored IEEE ISCAS 2026 · Co-authored npj Wireless |
| Year | Venue | Title |
|---|---|---|
| 2026 | IEEE ISCAS (Accepted) | ISAC for Intelligent Transportation: Ray-Tracing, Clutter Cancellation, and Sensing-Aided Beamforming |
| 2026 | npj Wireless Technology (Accepted, Nature Portfolio) | Integrated Communication and Sensing: Algorithms & 3D Simulation Insights |
| 2024 | EACL, DravidianLangTech Workshop | Fake News Detection Using Deep Learning Models in Dravidian Languages |
| 2023 | CEUR, FIRE 2023 | Sarcasm Detection in Dravidian Languages using Transformer Models |
| 2023 | ACL, LT-EDI Workshop | Transformer Models to Detect Levels of Depression from Social Media Text |
Open to AI Engineer · ML Engineer · NLP Engineer · Data Scientist roles in Singapore Available from July 2026 · View Portfolio →