NonclinicalVial · drug development
Solving preclinical data organization
Vendor spreadsheets in; flagged, traceable data and one-click IC50 curves out.
→Read case studyVP of Product, Vial · ex-oncology research coordinator
Vial's first product hire, now VP of Product running product, engineering, design and research. I led the team that built Vial's clinical trial software from scratch: the EDC, randomisation, and the portals that sites, monitors and patients use. Then the AI that works on top of it. Before Vial I coordinated CAR-T and immunotherapy trials at UCSF and MGH, which is where most of my opinions come from.
Currently · VP Product at Vial
🩺 bedside → build
APM → VPin four years, as Vial's first product hire
5 appsbuilt from scratch: EDC, RTSM, Start-Up, Monitoring, Admin
~4.7×cheaper than vendor EDC + RTSM across five Phase 1 trials
60 → 92%catch rate of the AI monitor I own, over six iterations
What kind of PM am I?
Three things I hold myself to.
01
How you communicate, 90% of the time, is the job.
Nobody experiences a PM through the strategy deck. They experience the 4pm Slack reply, the ticket, the tradeoff explained to a coordinator who has to live with it. I write clearly, early, and to the person who has to act.
02
Does this meaningfully support the business?
I ask it about my own work more than anyone else's. If the honest answer is a shrug, I stop, say so, and move the effort somewhere it counts.
03
First principles keep that answer honest.
Start from what the trial, the site or the company actually needs. Not from the roadmap, and not from what a vendor happens to sell.
The story · bench → bedside → build
One path through the pipeline, told in the products.
OriginUCSF · MGH
It starts at the bedside.
Oncology research coordinator on CAR-T and immunotherapy trials. Every product decision since has been a vote for the person entering the data, and for changing how core healthcare is delivered, industry-wide. That is the work I have given my life to.
CAR-T & immunotherapy trialsHeme-Onc
01
Chapter 01
Nonclinical — the bench
Before a drug reaches a patient it generates thousands of assay rows, and nobody has time to clean them.
02
Chapter 02
Clinical — the trial
Where the software is regulated, the people are tired, and every field exists for a reason.
03
Chapter 03
AI tooling — the layer
AI that touches trial data has to earn it. Mine keeps a person in the loop until it does.
OriginUCSF · MGH
It starts at the bedside.
Oncology research coordinator on CAR-T and immunotherapy trials: consenting patients, chasing queries on Friday evenings, learning what a badly designed form costs a site. Every product decision since has been a vote for the person entering the data, and for changing how core healthcare is delivered, industry-wide. That is the work I have given my life to.
CAR-T & immunotherapy trialsHeme-OncJiang 2016
01
Chapter 01
Nonclinical — the bench
Before a drug reaches a patient it generates thousands of assay rows, and nobody has time to clean them.
02
Chapter 02
Clinical — the trial
Where the software is regulated, the people are tired, and every field exists for a reason.
ClinicalVial · 2022 →
Building our own clinical trial software
Five Part 11 applications, built in-house, now running every Vial Phase 1 trial at ~4.7× less than vendors.
→Read case studyClinicalVial · data management
Automating clinical data management
Coding, queries and TMF tracking done by software. A person still signs every run.
→Read case studyClinicalVial · program management
Shipping regulated software every two weeks
A validation package ready to sign the day the code merges.
→Read case study03
Chapter 03
AI tooling — the layer
AI that touches trial data has to earn it. Mine keeps a person in the loop until it does.
AI toolingVial · AI product
Using AI to review trial data
An agent drafts the findings and a monitor approves them. Catch rate 60% → 92%.
→Read case studyAI toolingVial · AI product
Drafting regulatory documents with AI
IBs, protocols and Pre-IND packages assembled from program data, every section cited.
→Read case studyHow I work
Forward-deployed, by default.
I don't hand problems to a backlog. I sit with the coordinators and monitors, then build. Most of what I build is in production the same month.
01
Forward-deployed.
I'm in the standup, I read the queries, and the fix ships that week. The roadmap comes from the site's problems, not from a slide.
02
Human-in-the-loop until earned.
Regulated AI starts as an assistant. I track the override rate like a product metric, and a system gets more autonomy only when the evidence says it can.
03
Spec it so an agent can build it.
A ticket precise enough for an AI agent to implement is precise enough for anyone. One engineer reviews, and the team's hours go to judgement.
04
Validation is product.
A release isn't done until the validation package is ready to sign. Documentation that lags the code is a bug.
build log · things I shipped, not slides6 tools
clinical-monitor
AI remote data review over EDC data
medical-coding-automation
Bi-weekly MedDRA / WHODrug pipeline
ib-generator
22-prompt Investigator's Brochure drafting
pind-generation
16-section Pre-IND meeting packages
pbmc-assay-processing
CDMO assay files → QC-flagged, Scispot-ready
deployment-docs
FRS, trace matrix and OQ updated per release
The non-proprietary pieces are being cleaned up for GitHub.GitHub · coming soon
I do these myself:SQL: I pull my own data (Metabase, Postgres)Python data pipelinesReact prototypesn8n automationsClaude skills & agentsFigma → FramerLinear, written like a spec
A Tuesday, with AI.
How it shows up in a day. A habit, not a strategy slide.
The daily market briefing is already in Notion. A skill wrote it; I read it.
Standup. Two tickets specified tightly enough that a coding agent builds them and one engineer reviews by lunch.
Someone asks about enrolment. I write the SQL in Metabase and answer in the thread, with the chart.
The AI monitor has drafted this week's findings. I review the overrides, not the records.
Prototype a payment-milestone flow with Claude. The screenshot goes to ops and design to decide, not to debate.
A skill drafts the weekly readout from Linear and the deployment log. I edit the judgement and send it.
AI drafts. I decide. Anything regulated keeps a human signature.
Off the clock
Personal projects, mostly handmade.
Side venture
Sour Cherry
A gifting brand for Indian weddings and festivals — welcome bags first, Holi and Diwali next.
Textiles
Brishti
Silk headscarves. Monsoon-named.
Poetry
Sunday Morning Tabla
A collection, written in the long shadow of Tagore and Rumi.
Suno · Ableton
Generative music
Experiments in letting a model hum and a human arrange.
Let's build something regulated and delightful.
Looking at Director, Head and VP of Product roles in clinical-trial software, digital health and life-sciences tooling. Remote in the US, New York or the Bay Area.
© 2026 Dhruba Biswas · built by hand.Bench → bedside → build.