A platform that turns news into ranked content before competitors notice the story.

Why this exists

How it all fits together

Four worker services, one approval service, Redis streams between them, Postgres for state. Click any component to learn what it does.

The 4-gate cost funnel

The most important architectural decision in the platform. AI calls are expensive. Every article passes 4 gates. Each gate is cheaper than the next, and drops most of what enters.

At $99/month per client, AI cost is 2.6% of revenue. Without the funnel: ~$450/client/month. Catastrophic.

Source spread × Google Trends

Gate 3 computes two independent signals before deciding to generate. One comes from our own data, one from outside. Drag the controls to see how they combine.

The 8 angles · how we differentiate

Competitors rewrite news. We generate opinionated content with a carefully selected angle. Claude is given the full list and picks the best fit for each article + client combination. Click each card to see how it sounds in production.

Competitor gap analysis

Before each Sonnet generation call, we check what competitors have already published on this topic. Three cases, three different prompts to Claude.

Multi-tenancy & data isolation

Row-level isolation. Every table with client data has client_id UUID. client_id is never trusted from the request body — it comes from the signed JWT only. One bug cannot leak another tenant's data.

How we ship

Two paths. kind for local development, ArgoCD + GHCR for production. Same Helm chart, different values file.

kind create cluster --name content-intelligence
kubectl cluster-info --context kind-content-intelligence

docker build -t ci/ingestion-worker:dev -f services/ingestion-worker/Dockerfile .
docker build -t ci/intelligence-worker:dev -f services/intelligence-worker/Dockerfile .
docker build -t ci/approval-service:dev -f services/approval-service/Dockerfile .
docker build -t ci/publisher-worker:dev -f services/publisher-worker/Dockerfile .

kind load docker-image ci/ingestion-worker:dev --name content-intelligence
# repeat for the other three

kubectl create namespace content-intelligence
bash scripts/create-secret.sh

helm upgrade --install ci ./helm/content-intelligence \
  -f helm/content-intelligence/values.yaml \
  -f helm/content-intelligence/values-prod.yaml \
  --namespace content-intelligence

kubectl get pods -n content-intelligence -w
kubectl port-forward -n content-intelligence svc/ci-approval-service 8000:8000

# then open:
#   http://localhost:8000/register
#   http://localhost:8000/admin

git push -u origin master
# GitHub Actions builds all 4 images, ~8 min first run, ~2 min after

PUSH_TO_SSM=true bash scripts/create-secret.sh
# writes /content-intelligence/prod/platform/{key} in SSM

helm install csi-secrets-store \
  secrets-store-csi-driver/secrets-store-csi-driver \
  --namespace kube-system \
  --set syncSecret.enabled=true

kubectl apply -f https://raw.githubusercontent.com/aws/secrets-store-csi-driver-provider-aws/main/deployment/aws-provider-installer.yaml

kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml
kubectl apply -f argocd/application.yaml
# ArgoCD syncs the Helm chart automatically

# in Cloudflare Zero Trust → Tunnels → Public hostname:
# subdomain: content
# domain:    pratheesh.com.au
# service:   http://ci-approval-service.content-intelligence:8000

Where we are · where we're going