About — Zeebrafish AI

15 yrsEnterprise finance & data experience

Kaiser PermanenteNational Medicare Finance, PARR

The Walt Disney CompanyCorporate financial systems

1Founder who's lived the problem daily

The founder

Siam Kazi

Siam Kazi is a Data Reporting and Analytics Consultant on Kaiser Permanente's National Medicare Finance team, working within the Payment Adjustment Reconciliation and Reporting (PARR) function. The work is exactly what it sounds like: validating CMS payment data, supporting SOX control processes, and producing CMS/MBE reporting and risk adjustment analytics across every KP Medicare market.

In practice, that means living inside MMR files, PPR/PRS reconciliation, HCC risk adjustment, the V24/V28 model transition, and Part D payment data — the exact formats Zeebrafish AI was built to decode. This isn't a product built by outsiders trying to reverse-engineer what Medicare finance teams need. It's built by someone who was asked those questions, in that seat, this year.

Before moving into PARR, Siam spent eight years on Kaiser Permanente's Health Plan Product, Service & Administration team, leading market strategy analysis that reached Director, VP, and SVP-level stakeholders — and, on occasion, KP's most senior executive leadership. And before Kaiser Permanente, he worked as a Senior Financial Systems Analyst at The Walt Disney Company, inside Disney's Corporate Financial Systems group, supporting BI planning and reporting for Corporate, Studio, Consumer Products & Interactive, and ABC TV on SAP HANA, SAP BW, SAP BusinessObjects, and IBM Cognos TM1.

165 → 6 hrs

Reduced Disney's $54M Guaranteed Shortfall reporting cycle from 165 hours to 6 via automation.

16 hrs/qtr saved

Automated Disney Consumer Products' $5.4M true-up/true-down licensing process.

That pattern — finding the manual, high-stakes, poorly-tooled financial process and automating it — runs through the entire career. Zeebrafish AI is the same pattern applied to the biggest, least-automated manual process in Medicare finance: reading CMS's own data.

Why Zeebrafish exists

Built out of frustration with a problem that shouldn't still exist.

Every month, CMS sends health plans payment and membership data in file formats that predate almost everyone currently working with them — fixed-width layouts with no headers, cryptic field codes, and a 200-page technical manual as the only documentation. Understanding them is a skill, not a tool. And it's a skill that lives almost entirely in a handful of people's heads.

Working inside PARR made that gap impossible to ignore. The people who could fluently read an MOR or a PPR/PRS file were often the most senior — and closest to retirement. Junior analysts were left re-deriving field meanings by trial and error, or waiting on a colleague's availability. Consultants filled the gap at hundreds of dollars an hour, for work that, once you know the format, takes minutes.

Zeebrafish AI was built to put that expertise into software instead of letting it walk out the door with the next retirement. It's the tool Siam wished existed the first time he had to decode a flat file with nothing but a printout of the CMS record layout and a highlighter.

What Zeebrafish is setting out to do

Give every Medicare and Medicaid finance team direct access to the expertise their best analyst has.

The mission is straightforward: no health plan finance team should be one retirement, one vacancy, or one $400/hour invoice away from being unable to read the data CMS sends them every month. Zeebrafish AI's goal is to become the standard first step in every CMS reconciliation cycle — the layer between raw government data and a finance team's actual decisions.

That starts with plain-English explanation of any CMS flat file, and extends into the Payment Integrity Engine — independently recomputing what CMS should have paid and flagging exactly where it didn't — natural-language querying over financial data, and eventually direct connections into the data warehouses (Databricks, Snowflake) health plans already run.

Practitioner-built, not vendor-built

Every prompt, every flagged anomaly, every recommended follow-up question reflects real PARR and Medicare finance workflow — not a generalist's guess at what healthcare finance teams might want.

A parsing engine underneath the AI

For MMR files, layouts are decoded deterministically, field by field, against CMS's own published record layout, before Claude ever narrates the data — so the numbers are grounded in verified structure, not a language model's best guess at a fixed-width file. Other CMS report types are next on the roadmap; until then they still get a plain-English read, just without the deterministic parsing layer underneath.

Payment integrity, not just readability

For MMR files, the Payment Integrity Engine — internally, Shadow MARx — independently recomputes expected Part C and Part D payment at the beneficiary level, across every CMS payment scenario (standard, rebate, basic premium, Hospice, ESRD, PACE, and MSP), and reconciles it against what was actually paid. It turns a reporting tool into a control.

The engine

Codename Shadow MARx

CMS runs MARx to calculate your payment. We run Shadow MARx to check it.

MARx is the CMS system that calculates what Medicare Advantage and Part D plans actually get paid — risk scores, bids, rebates, and adjustments rolled into one number per beneficiary, per month. It's the source of truth, and it's also a black box. Shadow MARx is Zeebrafish's independent, parallel recomputation of that same payment logic, built directly from CMS's own published formulas — the same kind of manual cross-check Siam runs by hand on the PARR team today, now built into software.

MARx — CMS

The government's payment engine. Calculates what you're owed. Opaque, batch-run, and not built for finance teams to interrogate.

Shadow MARx — Zeebrafish

An independent, transparent recompute of the same calculation — beneficiary by beneficiary, always explainable.

Live today for MMR files, tested against every CMS payment scenario. Two edge cases the engine can't fully resolve from the MMR alone are disclosed in full, by design, on our Trust & data handling page — we'd rather flag what we can't verify than guess.

We recompute your MMR payment independently and show you every beneficiary where actual does not equal expected.

I've spent fifteen years automating the financial processes nobody wants to own by hand. This year, I landed in the seat where Medicare's version of that problem lives — and Zeebrafish AI is what happens next. It's that same instinct, available to every finance team the moment they need it, not just the ones lucky enough to sit next to someone who already knows how to read these files.

— Siam Kazi, Founder, Zeebrafish AI

Want to see it work on your own files?

We're running a limited pilot program with Medicare Advantage and Medicaid MCO finance teams. Upload a real file and get a full analysis back within five minutes — no commitment, no setup.

Request a pilot analysis →