Infrastructure & AI Engineer · Singapore
Twenty years building systems that scale — from bare-metal datacenters to Kubernetes fleets running 7,000+ services. SRE, DevOps, platform engineering, and now AI infrastructure.
On the AI side: 8x #1 on the HuggingFace Open LLM Leaderboard. Author of UNA (Uniform Neural Alignment) and MGS (MultiGumbelSampling). Cybertron 7B ran on Cloudflare Workers AI for nearly two years as the only independent developer model in their catalog.
Open source contributor to Kubernetes ingress-nginx, Argo Rollouts, and Atlantis. 10,000+ tracked experiments in Weights & Biases. Building infrastructure that disappears into use and models that punch above their weight class.
Head of Infrastructure and Science · Xendit · since 2022
Spanish (Native) · English (Fluent) · Italian (Intermediate) · Chinese (Conversational)
Kubernetes · Helm · Terraform · AWS · GCP · Cloudflare · Go · Python · PyTorch · Transformers · ArgoCD · Argo Rollouts · Atlantis · GitOps · Datadog · Cilium · KEDA · CDN · CI/CD
Remote · Hybrid · TZ: CET · SGT · AU
EU · US · AU · SG
Spearheaded comprehensive overhaul of infrastructure and engineering culture. Orchestrated 7-figure cost savings. Managing 7000+ services across dozens of clusters and thousands of nodes. Pure IaaC state with zero-downtime EKS lifecycle management.
Single-handedly scaled AI capabilities from nascent to millions of users. Architected hyperscale Kubernetes infrastructure on GCP. Designed state-of-the-art Inference Platform with 400% efficiency improvement. Pioneered 'SuperBooga' broker-based solution for high-throughput generative inference.
Large scale computing of containerized environments with IaaC on high-availability systems. Kubernetes at large scale with ArgoCD, ArgoRollouts, and ArgoWorkflows. Contributed to mainstream Argo, Nginx Ingress, and Atlantis.
OnPrem to Cloud Migrations expertise. Automation specialist with CI/CD, Terraform, Python, DevSecOps. Observability expert with EFK/Prometheus/Grafana. Led team of 9 members across 2 squads.
First SRE on board helping the organization implement SRE practices. Toil reduction by Python. Implementation of Monitoring platform with Prometheus. Definition of SLO, Error budget, Monitoring Dashboards.
First line of response in the largest worldwide CDN, over 10k+ physical servers in 200+ locations. Troubleshooting Kafka, ZK, K8s, Mesos, Ceph. DDoS mitigation and performance troubleshooting.
Consulting services for container projects. Deployed Kubernetes, CoreOS, Rancher, and Swarm clusters from scratch in on-premises and cloud environments. Implemented Software Defined Storage solutions. Defined DevOps CAMS standards.
Engineered high availability and disaster recovery infrastructure models for Tier-1 applications across two datacenters. Pivotal role in SDLC CD/CI process, automating pipelines using Puppet and Chef.
Chinese Mandarin and English language studies. Independent freelance consulting work. Travel and cultural immersion across China.
Defining Networking, Unix, and Security architecture roadmaps. Supervised implementation of new provider solutions and defined best practices and security requirements. Lab design and performance troubleshooting.
In-depth application vulnerability scans. IDS/IPS devices administration. Checkpoint Firewall-1 National Cores administration. Migration of old circuits from Nortel to modern IP solutions.
Vulnerability tracking and risk assessment with countermeasures. DDoS mitigation, packet inspection, pattern discovery. Forensic analysis of detected intrusions. SIEM correlation rules design.
Linux infrastructure planning and execution. OS hardening for Linux & Windows. Developed centralized tripwire-like platform from scratch using Python & Expect/TCL. FortiGate and StoneGate firewall administration.
Migrations of Bare-metal AIX & Linux to virtualized environments with ESXi & vCenter. Administration of AIX LPAR pSeries big computing servers. SAN storage administration with FastT, Hitachi, McData, Brocade.
Administration of Apache and Application Servers with performance troubleshooting. Elaboration of monitoring scripts. Administration of corporate DNS, LDAP, TACACS+ and DHCP servers.
Platform troubleshooting and incident escalation. Maintenance of platforms, patching, users management. Support for RIMA Network circuits (ADSL). Administration of Radius ACL & Users.
Eight #1 positions on the HuggingFace Open LLM Leaderboard, competing with major tech firms and AI labs using original post-training techniques.
★ 8x #1 HuggingFace Open LLM Leaderboard
7+ PRs merged into mainline Kubernetes, Argo, Atlantis, and SurfSense — addressing real production-scale problems.
★ Merged PRs in Kubernetes, Argo, Atlantis & More
Compact neural vision codec and visual tokenizer — 16:1 spatial compression at 97.69% fidelity, under 150K trainable parameters, batches 6× 40MP images on a single RTX 4090.
★ 150K Params, 40MP Batching, 97.69% Fidelity
Classification, retrieval, and NLP tasks from LLM embedding geometry — only lightweight forward pass components, 28x faster than conventional transformers.
★ 93% Classification, 28x Faster Inference
Cybertron 7B v2 hosted as a first-party model in Cloudflare's Workers AI catalog — the only third-party fine-tune in the lineup, served at the edge for nearly two years.
★ Only Independent Developer in Cloudflare's AI Catalog
An auxiliary loss-based architecture patch for HuggingFace Transformers, applied during SFT/RLHF. 18 public releases across multiple base models, with multiple #1 leaderboard positions.
★ 8 Public Releases, Multiple #1 Positions
A spec-driven agentic ecosystem for long-horizon engineering on enterprise brown-field code.
★ Every task lands as a reviewed spec before it lands as code.
Regularization technique using Gumbel-sampled noise during SFT/RLHF. Combines with UNA (UNAMGS) for additive performance gains.
★ Compatible with UNA for Additive Gains
Exploratory parameter-efficient adaptation that competes with LoRA on GLUE at <0.25M trainable params, with zero-overhead expert switching at inference.
★ Promising Early Results — More Research Underway
Redis-first, event-driven workbench with swarm intelligence for long-running Claude coding sessions. JSONRPC + WebSocket + MCP. Open source under MIT.
★ Anticipated Anthropic's Harness Pattern
Each source file becomes an autonomous Claude agent communicating via MCP — surfaces contract mismatches and assumption bugs through OpenTelemetry traces.
★ Open Source · Claude Agent SDK + MCP
Five custom datasets across math, knowledge, and RLHF — used in #1 leaderboard models and SingleMoM expert composition experiments.
★ 5 Public Datasets · Powering #1 Models & RLHF Experiments
Over 10,000 documented experiments in Weights & Biases — sweeps, ablations, and training runs underpinning every published technique.
★ 10,000+ WandB Tracked Experiments
Two decades of building in public — from glFTPd community tools in C/TCL/SQL in the early 2000s, to performance-first Docker images in 2016, to neural-net debuggers, admission mutators, and smart-home IoT today.
★ 20+ Years Shipping — glFTPd, Docker, K8s, ML, IoT