Product Manager · Clinical AI · Elsevier
AI-native product manager with 10+ years shipping enterprise SaaS and clinical AI at global scale. I obsess over one thing: giving clinicians instant answers without asking them to give up the rigour that medicine demands.
A GenAI clinical decision-support engine that gives clinicians a trustworthy answer in seconds — grounded entirely in peer-reviewed Elsevier content, with every claim traceable to its source.
Live at ai.clinicalkey.com →Medical knowledge is expanding faster than any human can keep pace with. At the point of care, a clinician has seconds — not hours — to find a reliable answer, and the stakes could not be higher.
Generic AI can produce a confident answer instantly. But strip the problem to first principles and the bottleneck was never intelligence — it was trust. If a clinician can't see where an answer came from, they can't act on it. The real question was never "can AI answer clinical questions?" It was "can AI answer them in a way a clinician is willing to stake a decision on?"
ClinicalKey AI answers only from trusted, peer-reviewed Elsevier content — never the open web, never ungrounded model knowledge. Every response is a structured, cited answer where each claim links straight back to the source passage, tuned to the relevant patient population.
It runs on one non-negotiable rule: AI access never exceeds a user's content access.
Medical entity recognition, intent routing, and expansion — with the population filter applied.
Semantic and keyword search run together, fused by reciprocal rank fusion.
A cross-encoder scores and reorders candidates to the most relevant passages.
An answer is synthesized only from retrieved sources, with inline citations.
Each claim is checked against its cited source before the answer is shown.
An AI reading layer inside ClinicalKey that lets clinicians understand a 30–40 page chapter or article in minutes — by asking, summarizing, and verifying against the exact source they're reading.
Live at clinicalkey.com →A clinician, student, or researcher opens a book chapter or journal article that runs 30 to 40 pages. Most of the time they don't want an AI to invent an answer for them — they want to understand the source itself, faster, and read it in a better way.
From first principles: the friction has moved. Search is brilliant at helping you find content — but it does nothing once you're there. The unsolved problem is everything that happens after you land on dense, trusted material: scanning it, interpreting it, and deciding whether it's even worth a full read — all without giving up the ability to verify.
Reading Assistant is a document-specific AI embedded in the content page. It summarizes the chapter, answers natural-language questions about that exact article, and links every answer back to the passage that supports it — highlighting the source text so verification takes a click, not a hunt.
It's deliberately narrow: it helps you understand what's in front of you, never improvises beyond it. Where ClinicalKey AI asks across the library, Reading Assistant asks about the chapter you're reading.
An AI overview of the chapter or article, so you can judge relevance and orient before committing to a full read.
Natural-language Q&A about the current source — "What are the main complications discussed?" — in your own words.
Every answer points back to the exact source passages that support it — the core trust mechanism.
The supporting passage is highlighted in the source, so verification is a glance, not a hunt.
Thumbs up or down on every response — turning usage into a signal that keeps answer quality improving.
Answers only from the content on the page — controlled, explainable, and safe for a clinical environment.
The document is already chosen by what you're reading — no corpus search needed.
Request a summary, or ask a plain-language question about this article.
The answer is generated only from this source's full text.
Every reference points to — and highlights — the exact passage.
One click confirms the answer against the source.
I don't start with a feature. I start with a real person's friction — validate the problem before the solution, prototype it in days with AI, then build the product around what the evidence actually says.
I go to the front line before writing a line of a PRD. Real clinicians, in real settings, telling me where the friction actually is.
I pressure-test whether the problem is real and worth solving before falling in love with an answer. Sometimes the data rewrites the entire thesis.
In the India point-of-care work, four clinical case validations showed international tools were clinically unsafe for Indian practice — in gestational diabetes, for instance, international guidance prescribes insulin-first while Indian guidelines prescribe metformin-first. And on ClinicalKey AI, finding that 74% of entitled users never asked a single question flipped the problem from "content gaps" to "activation."
I turn research into something clickable in days, using AI to compress weeks of design-and-build into hours — so stakeholders react to a working thing, not a slide.
I translate validated learning into PRDs, business cases, and go-to-market — and gate every release on evidence, not opinion.
Builds
From rapid, AI-built prototypes to live products in production. Filter by stage.
Took India's point-of-care product from an approved $1.8M business case (18-month discovery, 208 clinicians, 65% IRR) through PoC to Beta — with a localised content engine authored by 35–40 Indian specialists across 20 specialties.
GenAI clinical decision support answering across the ClinicalKey corpus — grounded in trusted content with inline citations.
ai.clinicalkey.com →A document-grounded AI reading layer inside content pages — summarize, ask, and verify against the exact source you're reading.
clinicalkey.com →A patented, AI-powered RT-PCR analysis SaaS deployed globally during COVID-19 — NHS, ICMR, Harvard Medical, CDS. $5.5M saved, 25% faster analysis.
A React prototype built in under 3 days from KOL research — department tracks, a GenAI chatbot, leaderboard, and ICU/dialysis simulation paths.
See the prototype →A React dashboard — activation funnel and query-density heatmap — that reframed the India investment thesis from content gaps to activation.
About
I'm a Product Manager at Elsevier (RELX Group), where I own the strategy for a clinical solutions portfolio used by clinicians in over 180 countries. My work sits where healthcare, generative AI, and real clinical trust overlap.
What defines it is going deep on both sides: owning production RAG pipelines and verification systems on the technical side, and building the business cases and go-to-market strategy on the commercial side — while prototyping fast enough to validate an idea before a single engineering sprint.
Every AI claim should trace back to a source a clinician can verify. Trust isn't a tagline — it's an architecture.
A working React demo in three days beats a deck in three weeks. I build to learn, then kill or scale fast.
A great clinical product fits how medicine is actually practised — which differs sharply from one country to the next.
Expertise
Track record
Own product strategy for a five-product clinical portfolio serving 180+ countries — ClinicalKey AI, ClinicalKey Physicians, ClinicalKey Now, and Reading Assistant.
Led the patented LTP RT-PCR Analysis platform deployed globally during COVID-19; $5.5M savings, 25% faster analysis, 30% lower latency.
Full product lifecycle for biotech research tools. Progressed from Product Trainee to Associate PM in four years.