PhD researcher at NUS, driven by one question: what is the true nature of reality, and can we reverse-engineer it? I develop unified theoretical frameworks at the intersection of mathematical physics and AI, architect multi-agent research systems with formal verification, and design proteins for immunotherapy.
A unified theoretical framework proposing that spacetime is emergent—arising from a Quantum Cellular Automata network governed by Category Theory and Von Neumann Algebras. Derives consciousness as a topological phase transition where computation achieves self-reference. 20+ interconnected papers covering Holographic Polar Arithmetic, Computational Action Principle, Resolution Folding, Riemann Ground State, and Standard Model field labeling. 17-book monograph series (~3000+ pages). All theorems trace back to axioms via DAG-based dependency chains with reproducible verification scripts.
Evoformer-based transformer architecture for antigen-specific TCR design with geometry-aware structural embeddings, Monte Carlo side-chain optimization, and biological constraint validation. CDR3β design pipeline completed with 36 synthesis-ready constructs for 12 targets. Research accelerated by 6 concurrent AI agents performing autonomous literature analysis, proof construction, self-critique, and formal verification in real-time across multiple fronts simultaneously.
Solved 6/10 problems (with substantive partials on 2 more) from the rigorous #1stProof benchmark (Abouzaid et al., arXiv:2602.05192). Produced a 129-page auditable paper with full Lean 4 formalizations and 478 verification scripts. Designed human-AI hybrid loops addressing LLM hallucination through DAG-based dependency chains tracing every theorem back to axioms.
Building 100 AI-powered open source projects in 100 days. Shipped AI Argument Judge (multi-party analysis with 5 judge styles, real-time streaming verdicts) and IELTS Story Adapter. Community-driven, rapid prototyping culture.
Full-stack application combining traditional Chinese fortune-telling with AI interpretation. Built with React, TypeScript, Gemini AI, Stripe, and Redis. Real-time generation, multi-language support, and quota management.
Language and generative models for protein design; AI-driven TCR engineering with deep mutational scanning and transformer architectures.
Protein language models for molecular chaperone variants in neurodegenerative diseases. Analyzed large-scale DMS datasets, developed models predicting compound mutation effects. Paper submitted to Nature Communications.
Designed 12 adversarial perturbation types for table reasoning robustness evaluation. Published at ACL 2023.
SNP embedding using SOTA language models and auto-encoders for correlation analysis and gene regulatory network graphs.
CNN-based personalized scoring function for high-precision GPCR target identification.
Single-cell RNA-seq bioinformatics analysis for epigenetics research. Co-authored paper in Journal of Biological Chemistry.
NER system with Dilated CNN + CRF + Bi-LSTM, trained on 200K+ emails, ~88% accuracy.