About
I'm Dimuthu Wickramanayake, an engineer in Singapore working on AI platform and agent infrastructure, with 7+ years running production systems on GCP and AWS. Most of my recent work is about making LLMs safe to point at real infrastructure: two Model Context Protocol servers my colleagues use daily for incident triage and data questions, and an internal assistant serving 50+ people with fail-closed data isolation and approval-gated writes. Underneath that sits the ordinary, unglamorous platform work that makes it possible — Kubernetes services holding 3,000+ RPS for 50M+ users, a multi-terabyte BigQuery platform, and CI/CD you can audit. I also publish open source, including Limbiq (an adaptive AI memory layer on PyPI), PII Guard (a Rust privacy proxy for LLMs), and platform-mcp.
This site is where I write things down properly — the architecture decisions, the postmortems, and the tradeoffs that only show up once something is in production. Everything is on the writing page, or inthe feed.
What I work on
- Platform & Backend Engineering
- High-throughput distributed microservices on Kubernetes with production API platforms, rate limiting, OAuth, and multi-tenant isolation.
- AI/ML Infrastructure
- LLM-powered workflows, inference pipelines, adaptive memory systems, LoRA fine-tuning, and privacy-preserving AI architecture.
- Cloud & Data Architecture
- Cloud-native on GCP & AWS with real-time streaming (Pub/Sub, Kafka, Kinesis), CDC with Debezium, and IaC with Terraform.
- System Design Consultation
- Scalable, resilient distributed architectures specializing in high-concurrency, fault tolerance, and event-driven patterns.
Elsewhere
The full work history lives on my CV.