Case studies · 8

Eight systems. All in production.

Each one is written the way I would want to read it: the manual process it replaced, the architecture, the thing that made it hard, the guardrail that made it safe to deploy, and what changed afterwards. Employer work is de-identified; freelance work is labelled as such.

Fig. 06⊘ It recommends. The engineer decides.

ProductionIncident Management / SRE

Incident Copilot

A ticket-triggered triage agent that pulls the right resolution procedure, the relevant logs and comparable prior cases the moment a ticket is created — then posts a recommended fix onto the ticket before an engineer has opened it.

Professional experience

→

Fig. 05⊘ Read-only by construction.

ProductionObservability / SRE

Observability Copilot

Natural-language access to logs, metrics, traces and live endpoint health across a multi-application Kubernetes estate — built on a custom Grafana-stack MCP server I wrote because the one I needed did not exist.

Professional experience

→

Fig. 03⊘ Agents write and check. People merge.

ProductionSoftware Engineering / Multi-Agent

Autonomous SDLC Pipeline

An issue-to-production loop run by cooperating agents. A Coder agent writes and checks the change, a Reviewer agent validates the PR, a person merges, GitHub Actions deploys to AWS — and when a deploy fails or production misbehaves, a Deployment or Production agent picks it up and feeds the fix back in.

hours → minutes

feature-to-deploy turnaround

→

Fig. 08⊘ The agent holds names, not keys.

ProductionAgent Infrastructure / MCP

Universal API MCP Server

One MCP server that lets an agent call any HTTPS API — GitHub, Vercel, Notion, Linear, AWS — with credentials it never sees. No custom MCP server per service: if a token exists, the agent can use it, only on the hosts that token is approved for.

1 server

instead of one MCP server per API

→

Fig. 01⊘ No index means nothing to go stale.

ProductionEnterprise Knowledge Management

Company Knowledge Base Agent

An enterprise-wide agent that answers questions against Confluence through MCP — no vector store, no separate index, no sync job to go stale.

30 → 5 min

to find internal knowledge

→

Fig. 04⊘ This is the whole point of the system.

ProductionCloud Operations / Autonomous SRE

DevOps Diagnostics Agent

A GitHub-issue-triggered agent with SSH into a production EC2 host, a hard allowlist of non-destructive commands, and no authority to fix anything it finds.

Freelance project — live production instance

→

Fig. 07⊘ Schema validation is the contract.

ProductionIntelligent Document Processing

Multimodal Invoice & Document Extractor

Turns invoices, PDFs, scans and photographs into schema-validated JSON using multimodal extraction — replacing template parsers that shatter the moment a layout shifts. Manual effort per document fell from about ten minutes to under one.

10 → <1 min

manual effort per document

→

Fig. 02⊘ Nothing posts until I approve it.

ProductionContent Automation / Brand Systems

Content Automation Pipeline

I tell a Claude agent an idea; it lands in Notion. A Claude Code job in GitHub Actions writes the post, fact-checks it, renders an on-brand image and writes the draft back to Notion. I review it there — and once approved, it is scheduled to LinkedIn for Tuesday or Thursday.

2h → 15 min

per day of content production

→

Before the agents

+3 earlier production systems

Supply Chain Forecasting System92% forecast accuracy
WebAI — Website Assistant80% efficiency gain
ANPR Model Fine-Tuning11% → 88% on EVs

On the about page →

Get in touch

Building an agent that has to survive production?

I write about MCP design, evals and guardrails on LinkedIn most days. If you're working on something similar, I'm happy to talk shop.