Production Lab:
What It Actually Does
Ten live systems on self-hosted infrastructure, covering geospatial, AI agents, data engineering, analytics, and security.
What you get
Live case studies
Evidence
Operational metrics
Format
Interactive D3 analysis
Real-time operational metrics
Production telemetry pulled from Prometheus: throughput, latency, events, agent load, and SLA posture.
Source
Prometheus
Cadence
15s scrape
Scope
5 VMs
View
Live trend + summary
24-hour signal observatory
A disciplined D3 readout of throughput, event pressure, and orchestration load across the production estate.
Window
24 hours
Series
3 live signals
Annotations
Burst + surge
Read mode
Hover inspection
Five domains, ten deployed systems
A compact view of the operating footprint, core stack, and headline result for each domain.
Coverage
5 operating domains
Delivery
10 deployed systems
Standard
Measured outcomes
Geospatial
73% coverage gap closed
AI Agents
25 autonomous agents
Data Engineering
99.9% QA pass rate
Analytics
21.6K candles/day
Security
92.3% precision
Explore every deployed system
Filter the live portfolio by domain and jump directly into each production case study.
Status
Live production only
Cards
Impact + stack + health
Destination
Full interactive drill-down
Healthcare Coverage
Where should the next clinic go?
Finds underserved populations using drive-time isochrones. One analysis surfaced 30,000 residents outside 15-min coverage and identified optimal placement for 4 new facilities.
73%
gap closed
30K+
newly covered
$2.4M
cost saved/yr
Financial Crime Detection
Block the transaction before it clears.
Real-time scoring engine evaluates every payment in under 100 ms. 92.3% precision at 88.7% recall, with autonomous AI investigation agents that build case files from RAG-retrieved evidence.
92.3%
precision
<100ms
score latency
88.7%
recall
Revenue Leak Investigation
€2.1M recovered. Zero SQL written.
AI agents autonomously trace revenue loss through a Medallion data lake, identify a payment gateway 3DS2 failure as root cause, and produce a boardroom-ready investigation report.
€2.1M
recovered
0
SQL queries
4h
time-to-insight
Crypto Intelligence
Regime shifts detected before the chart moves.
Ingests 21.6K candles/day from Binance, runs Hidden Markov Model regime detection in real time. Identifies bull, bear, and sideways regimes with 78% accuracy at sub-50 ms latency.
21.6K
candles/day
<50ms
p99 latency
78%
regime accuracy
Neural Atlas
Your infrastructure as a living 3D map.
Interactive globe with real-time GeoServer WMS layers, Proxmox node topology, and service dependency graphs rendered as a spatial network you can fly through.
3D
globe view
Live
WMS layers
12
nodes mapped
Vietnam Fishing Dashboard
Every vessel tracked. Every zone enforced.
Geospatial monitoring of fishing vessel activity across Vietnamese territorial waters. Density heatmaps, route replay, and automatic zone-violation alerts for maritime authorities.
500+
vessels tracked
24/7
AIS monitoring
3
alert zones
Agent Orchestrator
16 agents. Zero human operators.
Self-managing AI fleet that monitors infrastructure, triages alerts, runs security scans, and optimizes cloud spend around the clock without manual intervention.
16
active agents
24/7
autonomous
<50ms
coordination
Autonomous Agent Fleet
25 agents. Self-healing swarm.
Full-scale AI fleet across 5 VMs — sentinel brain coordinates security scanners, infrastructure monitors, data validators, and cost optimizers with zero human intervention.
25
live agents
5
VMs patrolled
0
operators
Medallion Data Pipeline
Raw streams → gold-tier KPIs in 30 seconds.
Enterprise Bronze → Silver → Gold architecture with quality gates at every transition. 12-service stack ingesting event streams into 7 S3 buckets with full audit trail.
12
services
30s
batch interval
99.9%
QA pass rate
End-to-end data pipeline
A six-stage production path from ingestion to intelligence and delivery.
Ingest
Kafka + APIs
Process
PySpark + Delta
Serve
Dashboards + alerts + APIs
Full-stack technology radar
The production stack mapped by layer, adoption, and system dependency.
Layers
Infra · Data · Intelligence · Delivery
Nodes
28 technologies
Interaction
Hover to inspect usage
Production stack clustered by operating layer
The stack is presented as a quiet force-directed map so visitors can read platform structure without visual noise.
Layers
4 operating layers
Nodes
28 technologies
Interaction
Hover + drag
View
Clustered topology
How systems talk to each other
A mesh view of shared data flows, orchestration paths, and downstream dependencies.
Flow types
Events · Reports · Geo layers
Shape
Directional chord map
Interaction
Hover for dependencies
Directional system relationships, not decorative links
The chord view isolates where case files, curated data, geo layers, and orchestrated tasks move between production systems.
Systems
10 nodes
Connections
11 weighted links
Read mode
Arc + ribbon hover
Focus
Downstream dependency
Self-hosted bare-metal stack
A self-hosted stack on Proxmox, routed through Cloudflare and Nginx, with live operations visibility.
Host
Hetzner AX102
Runtime
Proxmox + Docker
Ingress
Cloudflare + Nginx
Ops
Live console
$ sdl status --all
CONTROL Proxmox cluster heartbeat .............. stable
DATA Airflow DAGs active ........................ 12
ML Model serving latency p95 .............. 48 ms
SECURITY Fraud signal stream .................. healthy
$ _
Want a system like this for your domain?
I build data and AI systems that are measurable, reliable, and ready for production.