Professional work · Overview

Summit EHR: one product, two codebases

I maintain the live Convex application for Summit EHR and contribute to its Medplum/FHIR replacement. This page treats them as one product, so the work is not counted twice.

Employer: Zikra Infotech
Role: Full-Stack Developer, on a team codebase
Product: Summit EHR, a multi-clinic healthcare platform
Status: Live application: maintained and extended. Replacement: in development, with work in staging.

The product

Summit EHR is a multi-clinic healthcare platform. It exists as two implementations of the same product: the live application, built on Convex, and a replacement built on Medplum and FHIR. I work on both. In the live application I maintain and extend existing modules. In the replacement I contribute new ones.

Maintaining software that clinics use every day changes how you work. You read the existing code before changing it, you find out which modules depend on each other, and you ship a fix without treating it as a reason to rewrite something.

What I work on

Live application (maintained and extended)

Patient intake and digital consent. Google review synchronization, the reputation workspace and notification delivery. Organization offboarding with export and deletion safeguards. Team administration. PDF generation for signed documents. Fixes that came directly from clinic feedback, such as registration, payer search, consent layout and date input behavior. I also maintain and extend the patient portal: OTP login, notification preferences and in-portal clinic replies.

Replacement (in development, work in staging)

Multi-clinic isolation, including navigation, administration and protection against stale results. A streaming clinical AI chat client. Public appointment booking. Scheduling. Debugging long-lived streams across the API, the runtime and infrastructure timeouts.

In progress or prototype

An intake and PDF overhaul is still in progress. The fax and referral screens are a prototype with dummy data, not yet connected to a backend. AI-drafted reply suggestions for reviews are not described here because I have not confirmed their release state.

Deep dives

Patient intake and digital consent: how a long form survives partial failure, and why editing a consent template must not change a document a patient already signed.
Reliable integrations: why a failed Google review sync keeps the last complete data set, and how notification retries avoid sending twice.

How I deliver

I split large features into small pull requests that a reviewer can read. For example, the reputation work shipped as about 13 focused branches, the AI chat as about 11 pull requests, and clinic navigation and administration as about 10. That describes how the work was delivered and reviewed. It is not a measure of business impact.

I also made several performance changes: removing serial work from staff authentication, reducing full chat re-renders for each streamed token, loading only the team profiles that are visible, and reducing calendar clock updates. I do not have before and after measurements for these, so I describe them as changes and not as results.

What I am not claiming

Release status reflects repository snapshots from mid-October 2026. I have not independently verified live deployments.

I did not lead the migration, and I did not build every subsystem alone. This is a team product.
I did not train or evaluate any AI model, and I make no claim about clinical accuracy.
I make no claim of regulatory compliance. Export and deletion work is about safe data handling in the product, not certification.
No patient data, source code or screenshots from the product appear on this site.