Finds what an industry is genuinely working through, rather than what was published about it.
Multi-Agent Marketing System
Two agents on one platform, with a person approving everything before it ships. One researches an industry and takes a position; the other plans a week of posts reviewed as a batch.
The problem
What it was like before
Summarising what other people have written is a solved problem and a worthless one. Anything that reads like a digest of other articles is ignored, correctly.
The useful output is a position — this is what is actually happening in this industry, and here is what we think about it. That requires research, clustering and judgement, and it requires a human to agree before it goes out.
How it looks
Two in, one gate, feedback out
Both agents converge on a single human review, and what is rejected steers the next draft.
What we built
What it does
Related material is grouped so a position can be taken on a theme rather than on a single article.
The output argues something. That is the point of it.
Lays out a week of posts as a coherent set rather than as seven unrelated items.
A person reviews the week at once, which is a realistic amount of attention to ask for.
Approval is not a gate that merely stops things; what gets rejected shapes what comes next.
Where it got to
In production
HumanApproval before anything ships
BatchA week reviewed at once
RoutedRejections steer the next draft
Our view
The approval step is a feature
It is tempting to measure an agent by how little human involvement it needs. For anything that speaks in your name, that is the wrong measure.
Batching the review is what makes the human step sustainable. Approving one post at a time is a job nobody keeps doing; approving a week at once is a meeting.
Built with
Tech stack
- LangGraph
- FastAPI
- Claude
- pgvector
- PostgreSQL
- Next.js
- Python
The work behind it
Delivered under Workflow Automation and AI Agent Development.
More work
Other things we have built.
Sanctions Screening Agent
Watchlist alerts arrive all day and most are false matches. The agent clears the obvious ones against a rule engine backed by an LLM and sends every disagreement to a person with the evidence already assembled.
Read itLaw-Enforcement Request Intake
Requests arrive as letters with scanned attachments. The agent reads the mailbox, OCRs the attachments, verifies identifiers against core banking, classifies the intent and routes it — with an audit trail behind every step.
Read itInternal Policy Assistant
Staff ask a policy question in plain language and get an answer drawn strictly from approved documents, with the source named. Nothing leaves the bank, and the assistant says so when the documents do not answer.
Read itCTR to goAML Conversion
A compliance team spent hours a week turning Currency Transaction Reports into goAML XML by hand. Parser, XML builder, schema validator and audit trail, behind a interface officers actually use.
Read itRegulatory Reporting Engine
A registry of every periodic return a bank owes. The engine pulls the data, builds each report against the mandated template, then schedules and submits it behind a maker-checker step.
Read itAutonomous Social Media Agent
A ReAct loop that researches, reads its own history to avoid repeating itself, writes a strategy memo, critiques its own drafts, publishes, and learns from every rejection.
Read itWhatsApp Booking Agent
Patients book on the number a clinic already advertises, at any hour, in the language they normally write. Only genuinely free slots are offered, and anything clinical goes to a person immediately.
Read itHave something like this?
Describe the process and we will come back with whether it is worth automating, roughly what it would take, and what we would build first — or tell you plainly if it is not a job for us.