Work Projects Case studies About Contact Resume ↓

Case study · DrDash

DrDash

Practice intelligence for NHS primary care

Product lead · 2026 · Built with a UK-based partner who runs sales and practice relationships

Visit drdash.co.uk ↗

30+

NHS data sources

8

management areas

3

levels of detail

~2 months

to live

The problem

The NHS publishes a huge amount of data about every GP practice in England, every month. Appointments, phone waits, workforce, QOF, patient survey results. It's all public, and almost none of it is usable by the people running the practice. It sits across more than 30 sources, on different release schedules, in inconsistent formats.

Practice managers end up with no clear view of where they're falling behind on access targets or leaving income unclaimed. The tools that exist are built for clinicians and analysts, not for someone with twenty minutes on a Monday morning. The 2026/27 GP contract added new access measures, and no free tool tracks them yet.

30+ NHS sources across six areas, each on its own schedule More than 30 NHS published sources, grouped into six areas shown as loosely stacked tiles, each released on its own schedule: Appointments monthly, QOF annual, Workforce quarterly, Patient Survey annual, ODS weekly, Population annual. 30+ NHS sources Appointments Monthly QOF Annual Workforce Quarterly Patient Survey Annual ODS Weekly Population Annual

The call I made

I didn't want to build another dashboard. Practices already have charts. What they don't have is something that tells them which number matters this month.

So I positioned DrDash around gaps rather than data: what's slipping, compared with what, and what to check. When I mapped the competition, Access & Demand was the only high-priority area with no free tool covering it. That became the wedge. I also split the product early into a practice product and a network product, because the people buying them have different jobs.

Gaps, not charts.

What I built

Live
DrDash data pipeline: 30+ NHS sources across six areas More than 30 NHS published sources, grouped into six areas, Appointments, QOF, Workforce, Patient Survey, ODS and Population, feed a pipeline that handles list inflation and failed releases. The pipeline feeds a metric registry that records source, level and confidence, which feeds eight management areas, which are shown at three levels of detail, L1, L2 and L3. 30+ NHS sources Appointments QOF Workforce Patient Survey ODS Population Pipeline handles list inflation and failed releases Metric registry source, level, confidence 8 management areas L1 · L2 · L3 DrDash data pipeline: 30+ NHS sources across six areas More than 30 NHS published sources, grouped into six areas, Appointments, QOF, Workforce, Patient Survey, ODS and Population, feed a pipeline that handles list inflation and failed releases. The pipeline feeds a metric registry that records source, level and confidence, which feeds eight management areas, which are shown at three levels of detail, L1, L2 and L3. 30+ NHS sources Appointments QOF Workforce Patient Survey ODS Population Pipeline handles list inflation and failed releases Metric registry source, level, confidence 8 management areas L1 · L2 · L3

A Python pipeline that pulls more than 30 NHS published sources together, across six areas: GP appointments, QOF, workforce, the GP Patient Survey, ODS reference data and population figures. It's built around the known problems in NHS data, like inflated patient lists and source releases that fail or arrive late.

Underneath it is a versioned metric registry. Every metric records where it came from, what level it can be attributed to, and how confident it can be. Those rules are built into the schema, so the product can't overclaim even as it grows. A finding only gets a status if it passes minimum sample sizes and NHS confidence intervals.

It runs on Next.js and Vercel, with Supabase in the London region behind an authenticated admin portal. I took it from competitive analysis to a live, demo-ready product in about two months.

Three levels of detail

Every finding can be followed from the headline down to the figures behind it.

L1 · See what matters first

One page shows how the practice is doing and what needs attention this month.

DrDash practice assessment for a demo practice, listing phone waits above the ICB, routine appointments slipping and urgent requests on target.
The overview: what needs attention this month, at a glance. Demo practice, illustrative figures.

L2 · Understand why

Each of the eight areas shows where the pressure is, with things to check.

DrDash telephone waiting times view for a demo practice, showing long waits by hour and observations to look into.
Understand why: where the pressure is, and what to check. Demo practice, illustrative figures.

L3 · Check the detail

The figures behind every finding, with the PCN, ICB and England alongside, and the source and period on each.

DrDash detail view for a demo practice, showing long phone waits by weekday and hour and a peer comparison table.
The detail, with the PCN, ICB and England alongside. Demo practice, illustrative figures.

Where AI fits

AI built most of this. I directed the build through an AI-assisted workflow with Claude Code. I wrote specs, audited the real state of the code before each change, and checked everything it produced against the data. That's how one product lead and a partner got a multi-source data platform live in two months.

In the product itself, I chose to get the rules right before adding AI. In primary care, a wrong number about a practice is worse than no number. An AI summarising data it can't verify would break trust with the first practice manager who checks it. So every finding is decided by rules first, with its source and confidence attached. The AI layer comes next, and its job is to explain findings the rules have already verified, not to generate new ones.

Live

Now: how it's built

Build loop with Claude Code A repeating loop of four steps: Spec, then Audit the real code, then Build, then Check against the data, then back to Spec. Spec Audit the real code Build Check against the data
How I build with Claude Code.
Roadmap

Next: in the product

Where the AI layer will sit in the product NHS published data flows into the metric registry, which records source, period and confidence. The rule engine decides each status. Next, on the roadmap, an AI layer explains the verified findings before they reach the practice manager. Rules decide. AI explains. NHS published data Metric registry source, period, confidence Rule engine decides status Roadmap AI layer explains verified findings Practice manager Rules decide. AI explains.
The AI layer only reads findings the rules have verified. It never sets a status.

What the demo taught me

In October 2026 we demoed DrDash live to GP partners and practice managers at an industry conference. The clearest feedback was about priority. Practices care most about how quickly clinically urgent requests are seen, and that view wasn't prominent enough. We're reworking the overview to put urgent access first. Feedback from real practices is now shaping the roadmap, not my assumptions.

Urgent access first.

What's next

Roadmap
DrDash roadmap A practice-facing app, so practices can use DrDash directly. Benchmarking that moves from comparing a practice with itself to comparing it within its network, then regionally and nationally. Optional practice-supplied data to go beyond what the NHS publishes. A network product for PCNs. Then the AI layer: narrative assessments built on the rule-based findings. Practice app Benchmarking self, network, region, England Practice-supplied data PCN product AI narrative layer DrDash roadmap A practice-facing app, so practices can use DrDash directly. Benchmarking that moves from comparing a practice with itself to comparing it within its network, then regionally and nationally. Optional practice-supplied data to go beyond what the NHS publishes. A network product for PCNs. Then the AI layer: narrative assessments built on the rule-based findings. Practice app Benchmarking self, network, region, England Practice-supplied data PCN product AI narrative layer

Built with

← Back to vibhormadan.com