Shylesh Monceey
Available — Dubai / Abu Dhabi senior design leadership roles
Product design  ·  Clinical research  ·  Real-world data

nSights — turning clinical research into a few minutes of work, not weeks.

A multimodal, agentic platform for clinical researchers: define cohorts in plain language, compare them with real statistics, read the evidence behind every claim, and quantify medical images — all over the industry's largest curated EMR dataset.

Agentic AI Cohort building Comparative analytics Augmented curation Imaging biomarkers
01  —  The Brief & Goal

Real-world data holds the answers. Getting to them takes too long.

Most of what a clinical researcher needs is locked inside electronic medical records — and most of that sits in unstructured notes that no query can reach. Building a defensible cohort, comparing it against a control, and tracing every result back to source evidence is slow, manual work that can run for weeks before a single insight lands.

The goal was to collapse that loop. Let a researcher go from cohort to patient to evidence in minutes — with criteria that read like clinical language, statistics that hold up to scrutiny, and every claim deep-linked to the note it came from.

Scope — what the platform had to do
02  —  The Process

Make unstructured notes structured, traceable, and answerable.

The hard part of the experience is upstream of any screen: a clinical note is free text, and a researcher needs it as discrete, addressable facts. The curation pipeline reads a note, finds its section headers, predicts where each sentence belongs, and resolves the ambiguous cases — so a later question like "what was the tumour size, and did it change?" can be answered against labelled structure rather than a wall of prose.

That structure is what makes the rest of the product honest: every highlighted entity, every statistic, every AI answer traces back to a specific sentence in a specific note.

Task flow — section tagging a clinical note
Identify sentences Get section-header candidates Predict with sentence tokenizer model Disambiguate unclear section tags
A clinical note broken into tagged sections — Chief Complaint, History of Present Illness, Past Medical History and Medications — above the four-step tagging flow.
Structured note output. A free-text note resolved into labelled sections — Chief Complaint, HPI, Past Medical History, Medications — the unit the rest of the platform reasons over.
03  —  The Outcome

One workspace: cohort, patient, statistics and imaging.

The platform pulls cohort definition, statistical analysis, evidence curation, AI Q&A and medical imaging into a single workspace — so the researcher never has to leave one tool to verify what another claimed.

The Patient Explorer with a longitudinal event timeline on the left and the Patient AI-Expert panel answering a natural-language question about tumour size on the right.
Patient Explorer + AI-Expert. A patient's events on a timeline, beside an AI panel that answers plain-language questions and cites the exact notes behind each answer.
Cohort Builder showing a structured criteria picker beside CohortGPT interpreting a natural-language prompt into clinical events.
Cohort Builder & CohortGPT. Pick clinical criteria directly, or describe the cohort in a sentence and let the assistant translate it into events.
A diabetes cohort A versus cohort B comparison across diseases, FDA-approved drugs and surgical procedures with patient counts.
Cohort vs. cohort. Two cohorts compared side-by-side across diseases, drugs and procedures, each with live patient counts.
Demographic comparison of two melanoma treatment cohorts — age distribution, sex and ethnicity — with rate ratios, Chi-Square and Cohen's D.
Comparative demographics. Pembrolizumab vs. nivolumab cohorts compared on age, sex and ethnicity — with rate ratio, Chi-Square and Cohen's D in view.
A Kaplan-Meier survival probability curve for two cohorts beside a paired distribution chart.
Survival & paired analysis. Kaplan-Meier survival curves with confidence bands, alongside paired distributions that reduce inter-subject variability.
Imaging biomarker tools — image classification on an MRI slice and structured annotation on a histology image.
Imaging biomarkers. Classification and structured annotation on MRI and histology — qualitative scans turned into computable, labelled metrics.
A radiology report view pairing patient and study metadata with the corresponding imaging series.
Radiology report view. Patient and study metadata read against the imaging series and instance-level detail, in one place.
A specialty-organised disease library — cardiology selected — with condition cards, grades and patient counts.
Disease library. Conditions organised by specialty, each card a starting point into criteria and a live patient count — a fast way in for researchers who don't yet know their cohort.
What it adds up to
Minutes
From cohort to patient to evidence — instead of weeks.
Billions
Clinical notes mined and labelled through augmented curation.
5 modes
Cohorts, statistics, curation, AI-Expert and imaging in one workspace.
Traceable
Every claim deep-linked to the source note behind it.
Case study

Want the walkthrough behind these screens?