AISuffer

Cases

What is built and running, in plain terms

This page lists real systems, with their status. Items marked as in progress or planned have no results yet and none are invented. Client cases will be added here only with the client's permission and measured numbers.

Running

The automations that publish this site

Running
Problem
A one person content operation cannot afford to spend evenings on formatting, link checks, SEO checks and deployment.
What was built
A test suite that checks every page before it ships (metadata, schema, links, English only), a keyword lab that caches paid research so nothing is bought twice, n8n flows for the scheduled collection steps, and a deploy that runs on every push.
What it shows
The same pattern we sell as AI automation: plain rules where a rule works, a model only where a step needs judgment, and a person before anything goes out.
AI automation →

Hermes Agent, self-hosted with permission prompts

Running
Problem
Most agent demos answer questions but never take an action anyone would trust in production.
What was built
Hermes Agent (open source, by Nous Research) self-hosted on my own infrastructure, connected to messaging channels, with a permission prompt before any risky action and MCP tools for the systems it is allowed to touch.
What it shows
How an agent is allowed to act, and how it asks before it does anything irreversible.
Agents we run →

In progress

Private RAG over my own documents

In progress
Problem
Generic chatbots cannot answer from private documents, and sending contracts to a public API is not an option for many teams.
What was built
A fully local retrieval pipeline (pgvector plus a local model on a consumer GPU) that answers with page level citations and never leaves the network.
What it shows
That grounded, cited answers over a large private document set are possible without a cloud API call.
Private RAG →

An agent that operates infrastructure over MCP

In progress
Problem
Chatbots talk. Operations teams need something that can inspect a system, spot the fault and fix it.
What was built
An agent wired to custom MCP tools that inspects containers, finds a service that is down, and restarts it, with a permission guardrail before any write action.
What it shows
An agent completing an operations task end to end, with the guardrail visible.
AI agent development →

Planned

Daily brief: collect, summarise, email

Planned
Problem
Monitoring and summarising sources by hand eats the first hour of every morning.
What was built
A scheduled workflow that fetches sources, summarises them with a local model, and emails a formatted brief.
What it shows
The reporting pattern: scheduled collection, a model for the narrative, a person for the review.
Reporting process →

LLM cost audit, worked example

Planned
Problem
Companies suspect they overpay for model APIs but cannot see where.
What was built
A public teardown: model routing, prompt slimming, caching, and the break even point for self hosting, with the arithmetic shown.
What it shows
The method behind the readiness audit, applied to a real bill.
AI readiness audit →

Want to see one of these live?

A short call to see if this is a fit. No pressure, no slides.

Book a scoping call