Common questions

Filos helps hospitals identify inpatient malnutrition risk earlier and gives dietitians clearer chart support so they can spend less time hunting through records and more time with patients.

Primary users are clinical dietitians and nutrition leadership. Physicians, nursing, CDI, IT, and hospital leadership sit in the surrounding workflow—Filos fits that shared care context and keeps clinical judgment with the care team.

Filos helps teams identify and prioritize patients who may need nutrition attention, gather supporting information from the clinical record, and help the nutrition service focus scarce time where it matters most.

Assessment, diagnosis, and treatment decisions stay with clinicians. Filos works alongside your EHR, documentation, and CDI tools. See how we describe that fit on our Approach section.

A screening system needs to identify patients who may benefit from a closer look without creating an unmanageable review burden. We approach that as a tradeoff between sensitivity and precision: how many relevant cases are identified, and how many flagged cases merit the team’s attention. Rather than offering only a binary flag, Filos provides a scaled risk score to help dietitians prioritize patients for review.

Our founders have worked through this problem in computer vision quality assurance, including systems that flagged product issues for human review and a hospital meal-tray review system. That experience taught us to evaluate the full confusion matrix: true positives, false positives, true negatives, and false negatives, rather than relying on accuracy alone.

For each hospital, we’ll work with the team to evaluate missed cases alongside the volume and usefulness of flagged cases. The aim is to identify the right balance for the hospital’s workflow, so dietitians can focus their time where it matters.

Filos analyzes the clinical record for signals that brief admission questionnaires can miss, and it captures new clinical information to update patients’ malnutrition risk scores throughout their stays. Which data elements and interfaces a given hospital uses are scoped with that partner during evaluation.

Filos screens for malnutrition risk from EHR-style clinical data and supports dietitian review of that information. The product has been built and tested on publicly available, validated, authentic datasets. We are seeking early hospital partners for an approved real-world evaluation.

And there is more on the way—capabilities we are not ready to discuss in public yet. 🤫

Timothée Kampouris and Zuntue Pommerich met at Blue Sky Robotics. They bonded when they were put in charge of rescuing a failing pilot, putting in about 440 hours over three weeks. That stretch taught them to work alongside frontline staff, notice what already worked well before digging into problems, and listen carefully so feedback shaped the solution. After that period, the executives were pleased with the progress they had made.

Timothée was Director of Solutions at Blue Sky, running multiple client projects at once; Zuntue was CTO, leading a team of about seventeen engineers and interns.

That way of working—earn trust on the floor, then build—is how they approach Filos today. Roles, education, and fuller bios are on About.

Before Filos, Timothée and Zuntue led hospital foodservice computer-vision work that verified trays against printed diet orders before trays left the kitchen—onsite with the teams running the operation.

At Mayo Clinic, they ran onsite visits, collected tray-image data, stood up camera and network setups, developed the computer-vision models, planned the pilot, and worked through the hospital’s third-party risk management (TPRM) process.

At UNC Health, they led evaluation and pilot preparation—onsite tray data collection, technical setup, and an evaluation-stage BAA. UNC cleared the engagement for PHI use in that evaluation context, and remaining insurance requirements were completed as the program advanced.

Yes. The founders completed a business associate agreement (BAA) with UNC Health and successfully completed third-party risk management (TPRM) review with Mayo Clinic—real institutional security and contracting gates. That experience shapes how Filos prepares for partner reviews today. Current Filos controls are summarized on Security.

Yes. The working product integrates with Epic and has been tested in that environment, alongside the broader EHR, documentation, and CDI fit described on our Approach section. Connection details for a hospital partner are scoped together during evaluation to match that institution’s systems and security requirements.

We started Filos as we learned how widespread malnutrition is and how often it goes unrecognized. As we learned more, a close family member experienced severe malnutrition. Suddenly, the problem we were studying was affecting someone we loved.

That experience strengthened our commitment to addressing the gaps in nutrition care. We want to help hospital teams recognize patients’ needs earlier and give them more time and support to act.

Filos was founded in September 2026.

Filos Technologies, Inc. is our legal entity. We are a Delaware corporation in good standing.

Filos means “friend” in Greek. At our previous company, a client told us he saw us as partners rather than vendors—and the way he worked with us made that clear.

His approach left a lasting impression on us. When we started our own company, we wanted to carry that same spirit into how we treated people and built relationships. We chose the name Filos as a daily reminder of the kind of business we want to run.

We’ve been struck by how much care teams accomplish with the time and resources they have. They manage enormous workloads, complex patient needs, and information spread across many systems. Even with that skill and effort, important needs can still be missed.

We want Filos to take on more of the searching, organizing, and repetitive work that fills their days, giving them more room to focus on patients.

The clinical idea—that hospital records can help identify malnutrition risk beyond brief admission screens—is grounded in published research by independent investigators. Our Approach section points to peer-reviewed work on using hospital records for malnutrition risk, including a PLOS ONE 2022 study of malnutrition trajectories (PubMed ↗).

Filos builds on that literature with modern AI methods; published studies remain attributed to their authors.

Filos has built evaluation work on public clinical and nutrition datasets, including MIMIC-IV–oriented protocol work comparing classical and gradient-boosted models on structured data available early in a stay. That work includes careful review of label quality, missingness, rare positives, units, and timing so model development rests on clean, well-understood data.

With early hospital partners, we define endpoints, review methods, and success criteria together up front, so evidence grows from shared evaluation.

Filos organizes the evidence—why a patient was flagged and what chart support was gathered—so dietitians and the care team can act with clearer information. They stay in the driver’s seat for assessment and care decisions, including when documentation conflicts or information is incomplete. Diagnosis and treatment orders stay with the care team.

We take that concern seriously. Our goal is not to maximize diagnosis rates or assign a billing code based on an AI prediction. It is to help care teams find patients who may have been missed, assess them, and document diagnoses that are clinically supported.

The evidence shows why both sides matter. In a 2024 study by Aktas and colleagues of hospitalized patients with cancer, 197 of 420 patients identified as malnourished by a dietitian did not have a corresponding physician-coded diagnosis. That is evidence of a documentation gap in one clinical population, not a national missed-diagnosis rate. Separately, a 2020 federal OIG audit found substantial overbilling in a targeted group of historical severe-malnutrition claims, including cases where the record did not support the severity or effect on patient care.

Our approach separates screening, clinical assessment, diagnosis, and coding. A risk score is a prompt for review, not a diagnosis. We intend for the clinical evidence behind a recommendation to be traceable to the patient record, for dietitians and the responsible provider to exercise independent judgment, and for coding decisions to remain subject to the hospital’s CDI and compliance processes. Unsupported diagnoses or severity should be corrected, not carried forward to increase reimbursement.

We plan to evaluate clinical validity and documentation quality alongside identification rates, including cases that should not be diagnosed or coded. Success means helping the right patients receive appropriate care, with documentation that accurately reflects their condition.

Our current product is designed to meet FDA’s criteria for non-device clinical decision support—a category of software that does not require FDA device clearance or approval (FDA guidance ↗).

Filos helps hospital dietitians prioritize patients for assessment using existing health-record information. Our design makes the supporting information and reasoning available for independent clinical review, while assessment, diagnosis and treatment decisions remain with the care team.

As Filos and FDA guidance evolve, we will review changes in functionality and intended use, seek appropriate regulatory expertise, and obtain any required clearance or approval before introducing capabilities that require it.

For a hospital evaluation, Filos uses clinical data the partner approves for that purpose, under the agreements and controls that partner requires. Public demos use synthetic inputs. Retention, deletion, and model-improvement terms are set with the partner in those agreements.

Website visitor privacy (contact email, no ad trackers) is described on Privacy. Hospital product controls are summarized on Security.

Filos is built for hospital security reviews from the start: HIPAA-oriented controls, encrypted transit, and secure cloud environments, with room to fit a hospital’s compliance plan (including SSO where required). We maintain vendor BAA relationships with major cloud and productivity providers under each provider’s stated scope.

Details and current posture live on Security. Prior hospital TPRM and BAA experience is covered under founders’ background above.

Implementation is scoped with each partner: which systems Filos reads from or writes alongside, what identities and environments are used, and what success looks like for an initial evaluation. We start from demos and approved data paths, then choose the interface approach—whether FHIR, HL7, Epic-related, or another route—with IT, security, and clinical owners before any live connection is turned on.

You’ll work directly with the founders, Timothée and Zuntue. We love spending time on site, getting to know people and learning from the teams who have been doing this for years.

We want to become familiar faces—people your team feels comfortable talking to, asking questions and sharing ideas with. Understanding the day-to-day work matters to us, and so does getting to know the people doing it.

Your team will help shape what we build. You know the small frustrations, the workarounds and the things that could make a real difference to your day. We want to hear about them.

We’ll build around your existing tools and workflows, and keep learning from your feedback. We want the kind of relationship where you can tell us openly when something is helpful, when it’s frustrating and when we’ve misunderstood what you need.

We hope it’s the beginning of a relationship that lasts for years. Getting to know your team and understanding your needs takes time, and we want to keep building on that understanding.

If the evaluation demonstrates value, we’d work together toward a pilot and explore how Filos could support other teams across your health system. As your needs change, we want to stay involved— listening, improving what we’ve built and finding new ways to be useful to the people we’ve come to know.

Our conversations with hospital leaders and nutrition professionals have reinforced that patients with malnutrition are being missed and that improving identification is an important problem. One chief medical officer we spoke with was especially passionate about the need to address it.

Care teams also need the capacity to assess patients, act on the findings, and follow through on nutrition care.

Those conversations continue to shape what we’re building and how we think about fitting it into hospital workflows.

We start by learning your nutrition workflow and goals, then move into an approved retrospective evaluation on partner data before any prospective use. Scope, success criteria, data permissions, and security gates are written down before kickoff. We are looking for early hospital partners and expect to work closely with dietitians and their leaders throughout.

What matters most is access to your data and access to the floor—so the founders can observe frontline workers, hear their loudest complaints, and tailor Filos to how your team actually works. We are seeking a partner passionate about making change with cutting-edge tech and impacting care in a real way.

Scope, success measures, and commercial terms are agreed with the partner before work begins. As an early partner, you work directly with the founders; we are open to aligning part of our fee with shared results.

For a benchmark-based view of economic opportunity, use the opportunity calculator on the homepage.

Yes. The calculator draws on reported results from AI-assisted malnutrition screening at two major health systems. Cedars-Sinai reported $10.5 million in annual financial impact. Penn Medicine reported $23.8 million in annualized financial impact, based on three months of results. The reported impact includes additional reimbursement and the value of shorter hospital stays.

We use the hospitals’ reported results to develop per-admission benchmarks, then scale those benchmarks to your hospital’s annual admissions. The calculator provides a broad, ballpark estimate of potential opportunity, not a precise forecast or results achieved by Filos. Actual results will depend on your hospital’s patient population, payer mix, workflows, and implementation.

The underlying hospital results were achieved with another company’s system, not Filos. The company behind those hospital results has publicly documented its technical approach in peer-reviewed research, including the model architecture, clinical inputs, how information is extracted from clinical notes, and validation methods.

Yes. The United States has more than 5,100 community hospitals. We estimate that roughly 125 hospitals currently use dedicated AI-assisted malnutrition identification tools, or about 2–3% of that broad universe.

Our estimate is a bottom-up assessment of publicly reported deployments and inferred adoption across commercial vendors and hospital-built systems. One vendor reported screening patients at more than 50 hospitals; we estimate additional deployments from other vendors and internal hospital programs, with an assumed adjustment for possible overlap. Mount Sinai has reported using its internally developed system across six hospitals.

Public deployment counts are incomplete, and most vendors do not disclose how many hospitals use this specific capability. Our working range is approximately 75–300 hospitals. This is a provisional estimate, not a national census. It covers live AI workflows that identify malnutrition risk and support clinical action, not conventional screening questionnaires or every AI product that can mention malnutrition. The hospital denominator comes from an earlier AHA survey, so the comparison is approximate.

Cedars-Sinai has deployed a specialized AI tool for malnutrition and separately adopted AI-powered diagnostic and documentation support, illustrating that hospitals can use multiple tools for different clinical needs.

Existing deployments demonstrate demand, while the size of the market leaves substantial room for first-time adoption. We’re building Filos to earn that adoption through clinical usefulness, straightforward implementation, and measurable impact.

Yes. Hospitals have already adopted comparable AI-assisted malnutrition screening commercially. One established company in this space has publicly reported closing multiple deals worth more than $500,000 in annual recurring revenue each. Cedars-Sinai has also described a commercial implementation of this type of technology at its hospital.

That gives us evidence that this is a problem hospitals are willing to pay to solve. Filos is building for the same need, informed by the same clinical research, with our own approach to product design, advanced AI, tailored implementation, and hospital workflows.

AI-assisted malnutrition screening is our starting point, which we chose for its promising early real-world results and almost entirely greenfield market. We’re developing additional capabilities to help nutrition teams act on that information and increase their capacity to care for patients.

Our roadmap combines published clinical research, advances from the frontier of AI, and what we’ve learned working alongside hospital teams. Several capabilities are in development that we’re keeping private for now. Their purpose is to further improve how we identify patients at risk and keep those risk assessments up to date by capturing new clinical information.

With hospital partners, we develop and clinically evaluate those capabilities. We plan to measure their clinical performance and practical impact: the usefulness of the information provided, the work required from clinicians, and how effectively teams can act on the findings.

Early partners will help shape that development around their teams’ needs. We intend to earn our place through capabilities that demonstrate meaningful value in actual hospital workflows.