Can AI-Generated Discharge Videos Actually Fix Healthcare's Literacy Crisis?
The ritual is depressingly familiar. A patient shuffles toward the hospital exit, paperwork in hand—dense paragraphs about drug interactions, wound protocols, follow-up timings. Most of it might as well be hieroglyphics. That 2003 federal study on health literacy, the one showing only 12 percent of U.S. adults have proficient health literacy, gets cited so often it's practically a punchline. Except the consequences aren't funny: preventable readmissions, botched medication regimens, billions burned on avoidable errors.
Enter Framewise Health, a two-person operation fresh from Y Combinator's Spring 2026 cohort. Their theory? Video—specifically, personalized clips synthesized from a patient's actual medical chart—might cut through the fog where laminated pamphlets have failed. No app store friction. No login screens. Just a text message with a link, sent before discharge.
The premise sounds almost too tidy: yank diagnosis codes, prescriptions, allergy flags, and discharge protocols straight from the electronic health record, then let AI stitch together a short video tailored to that patient's treatment plan, preferred language, even literacy level. But in healthcare tech, simple pitches often hit complicated realities.
Mining the Medical Record
Framewise builds its videos by hooking into Epic, Oracle Health (the platform formerly known as Cerner), Athenahealth, and other major EHR systems via FHIR R4—the interoperability standard that became essentially mandatory across U.S. health systems by late 2025. Once a clinician flags a patient for discharge, Framewise pulls the relevant data: what's wrong, what's prescribed, what to avoid, institutional care protocols. The system layers in national drug databases, generates a script, then synthesizes the video.
Every clip gets reviewed by a clinician before it ships, according to the company's site, which saw updates in early May 2026. The compliance checklist reads like a regulatory pitch deck: HIPAA-certified, end-to-end encrypted, audit-logged, signed AWS Business Associate Agreement, SOC 2 certification in progress.
The product supports north of 70 languages—a leap beyond incumbents like Wolters Kluwer's UpToDate Patient Engagement suite, which offers videos and leaflets in up to around 20 languages. Framewise also aims to dial content to reading level, though the mechanics of how demographic data translates to comprehension-adjusted scripts remain largely in the company's promotional materials. It's mass customization, in theory, without the mass.
What Happens After the Video
Here's where it gets more interesting, perhaps. Most patient education platforms—Get Well Network's 10 million annual touchpoints, Krames' video libraries, Elsevier's Health Library—operate on a push model: deliver content, track views, hope for the best. Framewise aims to flip that, at least according to their pitch. Every interaction becomes a data exhaust.
Say a patient watches the medication segment three times but skips wound care entirely. The system flags it. Care teams can then prioritize outreach based on actual engagement patterns, not educated guesses or blanket protocols. Tane Kim, the CEO and a self-described med school dropout, frames this as an "engagement data platform" that could stretch beyond discharge—clinical trial onboarding, adherence tracking for chronic conditions, post-procedure follow-ups for device manufacturers.
Kim's origin story, shared in the company's May 11 Launch YC post, leans on firsthand observation: "People nod and say they understand, then leave and do the opposite." Co-founder David Cui, who has computer vision research credits from Brown and a prior internship at a YC-backed voice AI startup, built the video synthesis engine. Whether watching hundreds of confused patients is sufficient R&D for a scalable healthcare product remains to be seen.
Crowded Space, Novel Angle
Patient education isn't exactly uncharted territory. Wolters Kluwer, Get Well (now folded into SAI Group), WebMD's Krames arm, Elsevier—they've all been selling video content, EHR hooks, and multilingual materials to hospitals for years. Many layered in generative AI features over the past year as large language models went from novelty to baseline expectation: chatbots, summarization, predictive nudges.
Press Ganey debuted an AI product for patient satisfaction surveys in February 2026. Healthcare IT News ran a segment in April on AI closing "the medication information gap." At the HIMSS conference in March, SMS-based engagement tools with identity verification and real-time texting showed up across multiple vendor booths.
What those platforms generally don't do: auto-generate per-patient videos from EHR records on the fly. They curate libraries. Clinicians assign content. Views get tracked. Framewise automates the synthesis step, treating each discharge as a unique artifact rather than a templated playlist pulled from a dropdown menu.
Whether that approach scales—technically, economically, operationally—is the unanswered question. The company hasn't disclosed pricing, customer pilots, or deployment timelines. Third-party databases hint at early funding (CB Insights lists a $500,000 convertible note, though the entry is unclaimed and lacks a press release), but the team remains just two people according to their YC directory listing as of May 2026. That's lean, even by startup standards.
Trust Takes Longer Than Technology
Framewise's marketing materials emphasize speed of integration—"FHIR-native" suggests relatively frictionless EHR connections—and regulatory readiness, though formal SOC 2 certification is still in the works. For hospitals staring down readmission penalties or patient experience score pressure, the pitch lands cleanly enough: better discharge education, delivered in the patient's language, with real-time feedback on comprehension gaps. For pharma teams running support programs or trial recruitment, the engagement telemetry offers a new lens on patient understanding.
The target buyers are spelled out explicitly in the Launch YC post: specialty clinic owners, hospital chief medical information officers, VPs of patient experience, pharma medical affairs teams. Kim even lists his direct email. It's the kind of transparent outreach that either signals confidence or desperation, depending on how generous you're feeling.
The broader question—and it's not a small one—is whether healthcare organizations will embrace AI-generated patient communication at scale, especially for something as high-stakes as post-discharge instructions. The technical infrastructure exists. The standards are in place. FHIR R4 is real, interoperability is improving, and generative models can synthesize coherent videos.
But trust—clinical trust, legal trust, operational trust—doesn't compile as quickly as code. Hospitals move slowly, often for good reason. A two-person startup promising to automate one of the most sensitive touchpoints in the care continuum will need more than a clever demo and a Y Combinator pedigree. They'll need proof that the videos work, that patients actually understand better, that readmissions drop, that the data doesn't introduce new risks.
Framewise is live and hunting for early customers. Whether they find them, and what happens when real patients start clicking those text message links, will determine if personalized video becomes a genuine tool in healthcare's literacy fight—or just another well-intentioned feature that looked better in the pitch deck.
