Lifeline Billing

The problem

Lifeline Billing Solutions is a physician-led medical billing and revenue cycle firm in Texas, founded in 2023, serving a small panel of specialty practices.

When the engagement began there was no technology to speak of. No claims system, no CRM, no marketing infrastructure, and a domain that did not resolve. The billing work itself ran on spreadsheets and email, which is survivable across a handful of practices and stops being survivable shortly afterwards.

We came in as sole technical partner, reporting to the CEO, with a mandate to build the whole thing from zero.

What we built

Five systems, at three different stages. Three are live, one is in pilot, and two are still in build — each is labelled below, because a list of five systems reads like five finished systems unless it says otherwise.

LifeLine ClaimsFlow — live, and the internal team works in it daily. The system the business runs on: a claims platform covering the full lifecycle, from upload through AR follow-up and denial management to paid in full. Four roles — administrator, manager, AR, and denial management — each see the work that is theirs and cannot alter what is not.

It is built on Flask, React, and PostgreSQL, to HIPAA-conscious standards. Before it went live it went through a structured production-readiness review, which asked three questions in particular: whether a claim can ever be in two states at once, whether money is represented exactly rather than approximately, and whether an action performed twice does the same thing as an action performed once. Those are the three ways a claims system loses money quietly, and they were worth more attention than the interface was.

The website — live. It started from a broken domain and no indexed presence. Rebuilt on Vite, React, and Tailwind, deployed to Vercel, and prerendered, so search engines receive finished pages rather than an empty shell they have to run JavaScript to read. The conversion path is a lead-capture assessment rather than a contact form, so an enquiry arrives with something attached to it.

mCodeEasy — piloted at a client facility. Surgeons write their rounding on paper. Somebody then reads that handwriting and types it into charge capture, which is slow and is also the point at which billable work goes missing.

mCodeEasy turns photographed rounding sheets into structured charge-capture workbooks. It reads in two passes: once per page to extract what is there, then once across the whole batch to reconcile the pages against each other and flag whatever disagrees. The second pass is the one that matters — per-page extraction on its own is confident and wrong too often to bill from.

It runs on AWS Bedrock, on HIPAA-eligible infrastructure under a business associate agreement, because for this data the alternative was not an option.

The marketing CRM and content pipeline — in build. An agent-orchestrated CRM, and a content pipeline that drafts weekly SEO articles into an admin review queue rather than publishing them. A person approves before anything leaves. Alongside it, a structured LinkedIn programme across the company page and the founder’s profile: content pillars, a cadence, and keyword clusters mapped to the commercial-intent searches a practice actually runs when it is looking for a billing partner.

Outbound lead generation — in build. An agent that sources and qualifies practices against the firm’s ideal-client profile, and feeds the ones that fit into the CRM. Qualification happens before a person looks, rather than after.

What holds it together

These are deliberately separate systems rather than one platform. ClaimsFlow holds the claims data and nothing else reaches into it. mCodeEasy produces workbooks that a person reviews before they go anywhere near a billing system. The marketing side never sees clinical data at all.

That separation is a compliance decision before it is an architectural one. The fewer systems that can see protected health information, the smaller the surface that has to be defended, audited, and explained — and the easier it is to answer honestly when somebody asks where the data goes.

The AI in this engagement drafts and extracts. It does not decide. mCodeEasy proposes a charge capture workbook and a person checks it. The content pipeline proposes an article and an administrator approves it. The prospecting agent proposes a practice and a human decides whether to approach it. Every one of those steps could have been closed into a loop with nobody in it, and none of them was.

Everything runs in Lifeline’s name: their accounts, their repositories, their infrastructure. If they ever wanted to take the work elsewhere, there is nothing holding it here.