Hero appHero app

A patient's outcome depends on their hospital's workflowA patient's outcome depends on their hospital's workflow

Survival rates differ significantly between global centers of excellence and local hospitals. Our thesis: close the outcomes gap by standardizing workflows and empowering oncologists to make the best treatment decisions.Survival rates differ significantly between global centers of excellence and local hospitals. Our thesis: close the outcomes gap by standardizing workflows and empowering oncologists to make the best treatment decisions.

A standard workstation for every Hero, but only 3% came backA standard workstation for every Hero, but only 3% came back

Before Hero shipped on mobile, I designed an EMR-style web platform that oncologists use to review complex cases: pathology, radiology, treatment science, and clinical impressions in one workspace. But it had a retention problem: Before Hero shipped on mobile, I designed an EMR-style web platform that oncologists use to review complex cases: pathology, radiology, treatment science, and clinical impressions in one workspace. But it had a retention problem: only 3% of onboarded Heroes ever came backonly 3% of onboarded Heroes ever came back..

Retention came from one feature: peer discussionRetention came from one feature: peer discussion

Across 10+ video interviews, Heroes consistently valued the "Discuss with" feature: bringing a peer in on a case for a second opinion, a scan read, or a transfer call. They were doing the same thing outside the product on local messaging apps, but those threads were Across 10+ video interviews, Heroes consistently valued the "Discuss with" feature: bringing a peer in on a case for a second opinion, a scan read, or a transfer call. They were doing the same thing outside the product on local messaging apps, but those threads were fragmentedfragmented, , slowslow (waiting on peers in another channel), and (waiting on peers in another channel), and unsafeunsafe (sharing patient information wasn't allowed). (sharing patient information wasn't allowed).

Bring peer discussion on-platform, with AI as the always-on peerBring peer discussion on-platform, with AI as the always-on peer

Three design calls anchored the translation, and they stack: structure the data first, let AI surface insights through chat on top, then shape it to how oncologists actually work, between cases and at the bedside.Three design calls anchored the translation, and they stack: structure the data first, let AI surface insights through chat on top, then shape it to how oncologists actually work, between cases and at the bedside.

Heroes open a case with the record already loadedHeroes open a case with the record already loaded

Oncologists used to paste patient info into their own AI chats just to ask a question. Hero pulls each case straight from the existing EMR into a structured data layer (pathology, radiology, treatment science), so chat can work across the full case the moment it opens. No upload, no setup.Oncologists used to paste patient info into their own AI chats just to ask a question. Hero pulls each case straight from the existing EMR into a structured data layer (pathology, radiology, treatment science), so chat can work across the full case the moment it opens. No upload, no setup.

Heroes can ask anywhere across the patient recordHeroes can ask anywhere across the patient record

Chat sits on top of every section of the patient record, so Heroes can pull it up from anywhere they're looking and ask in context. AI synthesizes insights up front, the way oncologists already consult peers, so Heroes can question, push back, and refine through conversation rather than structured forms.Chat sits on top of every section of the patient record, so Heroes can pull it up from anywhere they're looking and ask in context. AI synthesizes insights up front, the way oncologists already consult peers, so Heroes can question, push back, and refine through conversation rather than structured forms.

Heroes get AI answers backed by cancer specialistsHeroes get AI answers backed by cancer specialists

To make AI responses credible enough for clinical use, I partnered with eng to build an internal tool where in-house cancer specialists train our AI agents. Specialists review complex cases, correct outputs, and add their own insights, and that expert judgment feeds back as knowledge, prompts, and model weights, so every answer Heroes see is grounded in specialist review.To make AI responses credible enough for clinical use, I partnered with eng to build an internal tool where in-house cancer specialists train our AI agents. Specialists review complex cases, correct outputs, and add their own insights, and that expert judgment feeds back as knowledge, prompts, and model weights, so every answer Heroes see is grounded in specialist review.

The feature Heroes asked for most: one clinical history, authored by the whole care teamThe feature Heroes asked for most: one clinical history, authored by the whole care team

Peer discussion is what kept Heroes coming back, but one request surfaced in nearly every interview: a single, structured Peer discussion is what kept Heroes coming back, but one request surfaced in nearly every interview: a single, structured clinical historyclinical history. The hard part is that it is *multi-author but single-reader* — intake and nurses capture the raw record, specialists and oncologists add interpretation, yet one oncologist reads it under time pressure at every treatment cycle. On the web platform I designed a role-based history timeline where each entry carries who added it and when, so the oncologist reads a coherent chronology while the whole team keeps it current.. The hard part is that it is *multi-author but single-reader* — intake and nurses capture the raw record, specialists and oncologists add interpretation, yet one oncologist reads it under time pressure at every treatment cycle. On the web platform I designed a role-based history timeline where each entry carries who added it and when, so the oncologist reads a coherent chronology while the whole team keeps it current.

The same history, built to scan at the bedsideThe same history, built to scan at the bedside

On mobile, Clinical History is read-first: the same structured record collapsed into a timeline Heroes can scan between cases or at the bedside — prior lines of therapy, response, and toxicities at a glance. It feeds straight into chat, so Heroes can ask about any point in the history in context.On mobile, Clinical History is read-first: the same structured record collapsed into a timeline Heroes can scan between cases or at the bedside — prior lines of therapy, response, and toxicities at a glance. It feeds straight into chat, so Heroes can ask about any point in the history in context.

Hero retention rose from 3% to 15%Hero retention rose from 3% to 15%

Two metrics validated the redesign. Retention: the share of Heroes who came back within 7 days, which rose from 3% to 15%. Conversation engagement: depth, reply rate, and how many cases Heroes worked through in chat, confirming they weren't just opening the app but actively using it to think through cases. Shipping Clinical History — the feature Heroes requested most — reinforced both, giving Heroes a reason to return between treatment cycles. Throughout, I ran regular sessions with a power-user group of oncologists, validating information architecture, conversation quality, clinical accuracy, and workflow fit. (Side note! I also helped produce interview videos by hiring videographers.)Two metrics validated the redesign. Retention: the share of Heroes who came back within 7 days, which rose from 3% to 15%. Conversation engagement: depth, reply rate, and how many cases Heroes worked through in chat, confirming they weren't just opening the app but actively using it to think through cases. Shipping Clinical History — the feature Heroes requested most — reinforced both, giving Heroes a reason to return between treatment cycles.

Throughout, I ran regular sessions with a power-user group of oncologists, validating information architecture, conversation quality, clinical accuracy, and workflow fit.

(Side note! I also helped produce interview videos by hiring videographers.)

Due to an NDA, I can't reveal more information. Please reach out to me via email (glimesong@gmail.com) for further details.Due to an NDA, I can't reveal more information. Please reach out to me via email (glimesong@gmail.com) for further details.