Practical AI for the work behind value-based care.
We build tools that take the repetitive work off your team, inside the systems you already use.
A visit with no claim is money nobody is looking for.
Billing dashboards built on the claims themselves, and utilization counts that mean what they say.
01Revenue integrityMeasured
2,727
claims captured in the first history load
Source: Build notes, Oct 2026
357 → 2,154
conditions confirmed as captured for one health plan
Source: The bot's run logs and build notes, Oct 2026
12,334 → 1,350
ER claim lines that were actual ER visits
Source: Commit history and notes, Jul 2026
The list of visits that were seen but never billedBillers stopped typing their own numbersCount events, not claim lines
Patients on the queue before they walk in.
The front desk stops re-typing what the record already knows, and a board shows who still hasn't been called.
02Patient accessMeasured
1,192
visits pre-registered in two weeks, none left unrouted
Source: The bot's run logs and build notes, Oct 2026
204
missed-appointment cards worked by 8 employees in two days
Source: Build notes, Oct 2026
248 → 13
prescriptions open past 14 days, after the cleanup
Source: Build notes, Sept 2026
Patients on the queue before they walk inA call board that closes its own cardsPrescriptions that looked overdue but weren't
Every line shows where it came from.
An RPA bot reads what is already on file, so your team spends its time on what isn't. A person approves before anything reaches a patient or a payer.
RPA bot: robotic process automation, software that clicks through screens the way a person does. It is not AI.
03Quality and codingMeasured
1,405 of 1,409
patients matched to their EHR chart
Source: Build notes and the bot's run logs, Oct 2026
about 160
patients read per hour
Source: The bot's run logs, Oct 2026
29 of 29
fax-form test cases passed, including faded and tilted scans
Source: Form-filling bench run, 14 Oct 2026
Chart prep without the clickingCoding gaps, with the evidence attachedRequesting outside records by fax, one safe step at a time
If the number was typed in, it's already old.
RPA bots and scheduled jobs keep one live view current for every department. Slow queries get fixed, not worked around.
04OperationsMeasured
4 + 47
on-site RPA bots and scheduled jobs keeping the dashboard current
Source: Build notes, Oct 2026
414 / 40
answered calls on one busy day / hand-offs credited to coworkers
Source: Build notes, Aug–Sept 2026
22 s → 0.5 s
the claim-capture monitor, after it was sped up
Source: Build notes, Oct 2026
A live operations dashboard for every departmentThe list of visits that were seen but never billedOne view of every member, across three health plans
Measure first. Automate the repetitive part. A person decides.
1
Measure today.
Every engagement starts by measuring how the work is done today, so results can be verified, not just claimed.
2
Rules for what gets audited.
Clear rules handle anything a payer or auditor will review.
3
AI for drafting and sorting.
AI does the drafting, summarizing and sorting.
4
A person approves.
Nothing reaches a patient or a payer until someone on your team says yes.
Built to fit the systems your practice already uses, and to protect patient data at every step.
Who we work with
Value-based care practices
Primary care groups with Medicare Advantage, ACO and health plan contracts. HEDIS and Star measures, risk adjustment, monthly plan files. This is where we started, and where most of our work is.
Specialist offices
Less manual work in the front office and billing, and more use of the data you already have.
Hospitals
The same approach for larger teams: measure the process, automate the repetitive part, keep a person in charge of the decision.
Six areas where the work piles up
Quality and care gaps
Tracking HEDIS and Star measures, prioritizing outreach and preparing chart reviews.
An RPA bot matched 1,405 of 1,409 patients to their chart.
Source: Build notes and bot run logs, Oct 2026
Risk adjustment
Surfacing documentation gaps with the evidence behind every suggestion.
Every suggestion shows where it came from: plan claims, our own billing or the problem list.
Value-based care finance
Dashboards for membership, utilization, spending trends and contract performance.
One view of every member across three health plans.
Front office and referrals
Insurance verification, pre-registration, referral routing and patient queues.
137 patients pre-registered on the first day, with zero duplicates.
Source: Bot run logs and build notes, Sept–Oct 2026
Revenue cycle
Claim tracking, rejections, unbilled visits and re-bill worklists.
Billers no longer type their claim counts by hand; an RPA bot reads them every 30 minutes.
Source: Build notes, Oct 2026
Operations
Live dashboards and daily performance reporting for every department.
Department by department, numbers typed in each day were replaced with live data.
Who's behind it
Thiago Bandeira
General Manager and Head of AI at a primary care practice · Miami, FL
I'm Thiago Bandeira, General Manager and Head of AI at a primary care practice.
I build the AI agents, machine-learning tools, RPA bots and data systems the practice runs on, and I use them every day alongside the people they're for.
The open-source projects are the same method in public: rules where it's audited, AI where it drafts, a person where it decides.
- Education
- M.S. in Data Science and Artificial Intelligence, Florida International University
- Certifications
- AWS Certified Generative AI Developer (Professional); AWS Certified Machine Learning Engineer (Associate)
- Languages
- English, Portuguese, Spanish
- In public
- Five open-source health-AI projects, all on synthetic or public data
The method, in public
Five open-source projects, built on synthetic patients, with their evaluations and limits published next to the code.
hedis-care-gap-agent
A rules engine finds care gaps, AI drafts the outreach, and nothing goes out until a person approves it.
Demo-grade, HEDIS-aligned, not NCQA-certified. Every patient is synthetic (Synthea). Its perfect engine score on the test split shows that the code follows its written rules over 60 synthetic patients; it is not evidence of clinical validity or of robustness to data the rules never anticipated.
hedis-spec-copilot
Answers questions about public HEDIS-aligned measure guidance and cites the source passage for every factual sentence.
Answers summarize public CMS and NCQA-published documents. This is not the NCQA HEDIS technical specification, and not clinical, coding or billing advice. HEDIS is a registered trademark of NCQA; the project is not affiliated with NCQA or CMS.
clinical-agent-evals
A test harness for clinical AI agents where every number is measured, never typed in.
A library for measuring agents, not a clinical product. Its metrics describe performance on the gold sets it is given, which in this portfolio are synthetic.
fhir-feature-service
Turns FHIR patient records into model-ready features that can never see the future.
Synthetic (Synthea) data only. Version 1 binds to localhost on purpose and has no authentication or TLS; it is a foundation for the portfolio, not a deployable service.
HCC Coding Assistant
Clinical language in, validated ICD-10 codes and a CMS-HCC category out, with every number coming from official CMS files rather than the model's memory.
A simplified RAF model, stated as such in every answer: it ignores V28 hierarchies, interactions and demographic factors. Public CMS reference data only; not coding or billing advice.
Start with 20 minutes.
Bring one process that eats your team’s day. We’ll talk through what to measure first, and whether automation or AI fits it at all.