EmoGalaxy 2: Teaching NPCs to Feel with Unity ML-Agents
A deep-dive into my thesis project — training autonomous game agents to exhibit emergent emotional behaviour using Unity ML-Agents and reinforcement learning.
Architecting the Next Generation of Intelligent Systems.
I bridge the gap between autonomous Agentic AI workflows, scalable cloud infrastructure, and high-performance interactive interfaces.
Expertise
Three Pillars of Practice
End-to-end ownership — from LLM orchestration and cloud infrastructure to production-grade interactive web.
Agentic AI & Orchestration
Designing autonomous, multi-agent workflows (Microsoft AutoGen, LangGraph) and high-performance RAG systems that automate complex research and business logic.
Cloud & Distributed Infrastructure
Engineering resilient, zero-downtime backend architectures using Kubernetes, Docker, and Kafka to handle high-frequency data and heavy LLM workloads.
Spatial & Full-Stack Web
Delivering production-grade, immersive user experiences utilizing Next.js, Postgres, and WebGL/Three.js to make complex AI systems accessible and visually engaging.
Case Study 01
UBICOMP Multi-Agent Ecosystem
Autonomous multi-agent orchestration at scale
A research-grade ecosystem where autonomous agents dynamically load skills, coordinate tasks, and are fully observable from a single telemetry plane.Dynamic Skill-Loading Architecture
Built a runtime skill-loading system allowing agents to hot-swap capabilities without redeployment — enabling continuous autonomous evolution of the swarm.
System-Wide Observability with OpenLit
Integrated OpenLit to instrument every LLM call across the ecosystem, capturing latency distributions, token usage, and cost attribution at the agent level.
Stack
Microsoft AutoGen · LangGraph · OpenLit · Python · FastAPI · Redis
Case Study 02
Musical Institution ERP (Freelance SaaS)
End-to-end SaaS delivery, zero prior infrastructure
Took a music school from spreadsheets to a fully managed SaaS ERP — owning every layer from database design to production ops.Full DevOps Lifecycle Ownership
Sole engineer responsible for architecture, development, CI/CD pipeline, and production deployment of a multi-tenant SaaS platform serving music schools.
PgBouncer & MinIO for Production Scale
Deployed PgBouncer for database connection pooling under heavy concurrent load, and MinIO for highly-available, S3-compatible object storage of student media files.
Stack
Next.js 15 · PostgreSQL · PgBouncer · MinIO · Docker · Nginx
Case Study 03
Distributed SCADA System
Enterprise-grade data pipelines, IoT to cloud
Replaced a monolithic SCADA relay with a fully distributed, message-driven microservices architecture capable of handling thousands of sensor events per second.Message-Driven Microservices with Kafka & MQTT
Architected a fully event-driven pipeline where MQTT sensors publish to a Kafka broker, decoupling data ingestion from processing and enabling horizontal scaling of consumers.
Kubernetes-Orchestrated Deployment
Containerized every service and orchestrated on Kubernetes, achieving self-healing deployments, automated rollouts, and resource-isolated workloads for 24/7 uptime.
Stack
Apache Kafka · MQTT · Kubernetes · Docker · Python · TimescaleDB
The Lab — Digital Garden
View all posts »Deep-dives, experiments, and build logs. From ML-Agents game AI and procedural animation in Unreal Engine 5, to self-hosting stacks with Dokploy/Cloudflare and AI evaluation tooling in Python.
Procedural Spider Locomotion in Unreal Engine 5
How I built a fully procedural spider rig in UE5 using inverse kinematics, terrain raycasting, and a custom gait state machine — no animation clips required.
Self-Hosting Everything: Dokploy + Cloudflare Tunnels on a Home Server
My complete self-hosting stack — Dokploy as the deployment platform, Cloudflare Tunnels for zero-open-port exposure, and a wildcard subdomain setup that makes spinning up new services trivial.
Evaluating LLM Outputs at Scale with Python
A practical walkthrough of the evaluation harness I built to benchmark LLM response quality, latency, and cost across multiple models — using DeepEval, custom rubric scorers, and OpenLit for observability.
40+
Projects Shipped
30+
Technologies
5
Agent Frameworks
4+
Years Building
Ready to build something
extraordinary?
Whether you have a complex AI system to architect, infrastructure to harden, or an immersive web experience to ship — let's talk.