Anurag Jha
← All products

NeuroDiscovery AI · 2026 · B2B · Cohort building for pharma

NeuroTerminal

my Neuro360 prototype shaped its query experience

Describe a patient population in plain English; get a structured cohort, step by step counts, a feasibility rating and outputs a pharma team can use.

My roleBuilt the Neuro360 prototype in 12 days and own the portfolio roadmap it sits on. A product manager on my team runs it day to day.

12 daysto build the Neuro360 prototypePrototype repository, Apr 2026
19 May2026 public launchPress release
6M+patient journeys behind the platformCompany press release
Neuro360: a question, matching patients, the agent's steps and the evidence chart.

The challenge

Sizing an eligible population across millions of records took a data engineer and weeks.

What shipped

Plain language queries parsed into coded criteria, counts from pre validated queries, an attrition waterfall, and a hand off to NeuroAudit for audited review.

What moved

Launched publicly on 19 May 2026. My Neuro360 prototype went from first commit to demo in 12 days

A

Ask, parse, count, show

The query experience my prototype set.

  1. 01

    Ask in plain words

    Suggested questions help a new user start.

    Neuro360 empty state.
  2. 02

    The agent shows its steps

    The query is classified, turned into graph queries, and checked for eligibility, in view of the user.

    Query classified.
    Thinking trace.
    01 / 02
  3. 03

    Counts from real queries

    The model never writes SQL. It parses intent; validated queries do the counting.

    Matching patients, eligibility checked. Drug name redacted.
    Attrition from millions of records to a cohort. Drug name redacted.
    01 / 02
  4. 04

    Evidence a team can use

    Narratives, charts and sources for every answer.

    Answer with evidence.
    Charts.
    01 / 02
B

The pharma workspace prototype

Ask once, route everywhere: cohort, feasibility, journey and audit in one place. Mock data.

  1. 01

    Command center

    One question, routed to the right module, with the role's permissions shown.

    Command center. Mock data.
  2. 02

    Intent, cohort, feasibility

    Structured intent, then the cohort, then a feasibility funnel.

    Structured intent.
    Cohort count.
    Feasibility funnel.
    01 / 03
  3. 03

    Journey and audit

    A patient journey view and an audit trail of every step.

    Patient journey.
    Audit trail.
    01 / 02

How it was evaluated

  • Four layers: criteria parsing, count correctness against hand written queries, clarification quality, and report faithfulness
  • One strict end to end success rate with release blocking guardrails
70 → 90%target success rate, launch to six monthsCompany reported · Team quality spec, Jun 2026

These are targets, not results.

What I learned

  • A prototype is the cheapest spec.
  • Keep the model away from the numbers.

Proof

Launch press release, 19 May 2026.