Pre-visit summaries
The AI reads problems, meds and recent results from the EHR, and files a summary back to the chart before the visit.
Keep building in Lovable or Replit. Our HIPAA compliant app development adds what the builder can’t: connections to healthcare systems, AI-written code made production-grade, and an app that scales in your own cloud, where you can see everything.
An AI app without integrations is a demo. Production means it reads from, and writes back to, the systems your users already work in.
Map your integrations in the free planner
By Jitendra Choudhary, Co-Founder and CTO · Updated
Before a prototype on the builder After the same screens, connected and in production
Code in your repository from day one. One named lead and a weekly demo. Stop after any phase and keep everything. Ask a question first
Automation only works when it can read the record and write back to it. These are the AI features our clients ask for, and the integration each one depends on.
The AI reads problems, meds and recent results from the EHR, and files a summary back to the chart before the visit.
The AI drafts the request from chart data, checks payer rules and submits it, with a person approving each one.
New results arrive as they post. Abnormal ones are flagged and routed to the right clinician, with a full audit trail.
The AI checks claim status in bulk, explains denials in plain words and queues the fix for your billing team.
Lovable and Replit are great for building screens and flows fast. Keep using them. These three steps are where healthcare engineers come in, in this order.
Epic, athenaOne and eClinicalWorks connections, from vendor approval to sandbox to production. FHIR and HL7, labs, payers and X12 clearinghouses, and medical devices.
We use the same AI tools you do, plus Claude workflows that make AI-written code production-ready. You keep the speed. We add the guardrails.
Production architecture in your own cloud account, load tested and scale tested, with monitoring and alerts so you always know what is happening in production.
The builder gets you to your first users fast. Growth asks more of the same app. Move the slider to see what holds as built, what needs tuning, and where we add engineering.
An honest split of the work. The builder keeps doing what it does well. Each gap on the right is something our workflow solves. Tap one to see both sides. New to the tool? Read how to use Lovable in healthcare.
The builder does well Calls simple APIs and webhooks well.
Expert hands add Epic, athenaOne or eClinicalWorks app approval, SMART on FHIR login, FHIR R4 reads and writes, and a production go-live checklist for each EHR.
The builder does well Builds the screens for orders and claims.
Expert hands add HL7 ORU result feeds, X12 270/271 eligibility and 837 claims through a clearinghouse, with retries, dead-letter queues and daily reconciliation.
The builder does well Generates new features in minutes.
Expert hands add AI review on every pull request, tests that block bad releases, and a person who approves each merge.
The builder does well Writes working code fast.
Expert hands add Keys and business rules moved from the browser to a server API, a typed data model, and handover docs so any engineer can change it.
The builder does well Gives you a live preview link in one click.
Expert hands add Dev, staging and production in your own AWS, Azure or GCP account, written in Terraform as IaC, with HIPAA-eligible services only.
The builder does well Shows errors while you build.
Expert hands add OpenTelemetry traces, p95 latency, error rate and message lag on one dashboard, with alerts paged to your on-call team.
The builder does well Runs smoothly for your first users.
Expert hands add k6 load tests at 10x your users, query indexes, caching, connection pools and auto-scaling rules before growth arrives.
The builder does well Can store and show patient data.
Expert hands add A BAA at every layer, row-level access rules on the server, an audit trail of every read, and MFA for every admin, logs sent to a SIEM, and patient data scrubbed from logs.
The builder does well Keeps your data in a managed database.
Expert hands add Point-in-time backups, a restore tested on a schedule, and a written recovery plan with RTO and RPO targets.
Each one happens inside the tools you already use. They make the app safer today and easier to take to production later.
Ten workstreams, run in parallel where they can be. This is a typical 12-week plan. Your readiness review sets the real one, with a fixed quote.
Need ongoing help after go-live? See healthcare software maintenance and support, or read how we run testing and quality engineering. Our guides go deeper on infrastructure as code for HIPAA clouds and logging without exposing PHI.
The goal is production, not a demo. Every phase ends with something you can check, and Evaluate feeds the next goal.
Your security team will ask this first. Here is the answer we put in the contract. Full detail is in our Trust Center.
| Area | You | Nirmitee |
|---|---|---|
| Cloud account and hosting | Own it | Configure it |
| BAA with the cloud provider | Sign it | Set up covered services |
| Encryption keys | Hold them | No copies |
| Production patient data | Handle it | Never handles it |
| Code and repository | Own it | Write and review it |
| Logs, backups, monitoring | Own them | Build and alert on them |
| BAA with Nirmitee | Signed before any work that touches patient data | |
We use AI in our own engineering, with the same gates on every client platform we run. Nothing merges without a person. For the wider program, see our HIPAA compliance and readiness work.
No patient data in prompts, at every stepAudit trail records every step
- const ai = new OpenAI({
- apiKey: "sk-live-...",
- dangerouslyAllowBrowser: true })
- await supabase.from("patients")
- .select("*")+ // browser calls your API only
+ await fetch("/api/summary", {
+ method: "POST", body })
+ // server: key from secrets manager,
+ // row-level rules, audit logFrom each vendor’s own pages, checked Nov 22, 2026. Plans change, so we recheck them in your review. Before an audit, run our healthcare security checklist.
| Platform | BAA | What their page says | Source |
|---|---|---|---|
| AWS | Yes | Business Associate Addendum for HIPAA-eligible services. | aws.amazon.com |
| Google Cloud | Yes | BAA for covered Cloud and Workspace services. | cloud.google.com |
| Firebase | Some products | Covered for some products under the Google Cloud BAA. Check each one. | firebase.google.com |
| Supabase | Yes | Sign the BAA and enable the HIPAA add-on. Self-hosted is not covered. | supabase.com |
| Vercel | Yes | HIPAA BAA add-on for Pro teams. Enterprise on request. | vercel.com |
| Render | Yes | Sign the BAA in workspace settings, then enable HIPAA. | render.com |
| Netlify | Unclear | HIPAA listed among audits. No BAA terms stated. | netlify.com |
| Lovable | No public BAA found | No BAA or HIPAA terms on the pages checked. | lovable.dev |
| Replit | No public BAA found | No BAA terms on the pages checked. | replit.com |
| Bolt | No public BAA found | No BAA terms on the pages checked. | bolt.new |
Client names stay private. The published case studies carry the detail. For the AI platform, a reference call is available under NDA.
For a US behavioral health EHR: 837P, 835 and 277CA issues traced to root cause and fixed.
Read the claims case studyathenahealth and eClinicalWorks integrations, from vendor approval to production. Plus an Epic SMART on FHIR patient app.
Map your integrationsMedical devices into an offline EHR. Passed ABDM M1 to M3. 14x monthly visits.
Read the device case studyWe build and run it in production for a US company, with a deployment in Saudi Arabia.
Reference call available under NDAStarting from an idea, not a prototype? See healthcare MVP development. Connecting to an EHR next? Plan the integration.
Yes. Many teams keep designing screens and flows in the builder. We connect your repository to a production pipeline in your own cloud, so new work ships through review, tests and approval.
Yes, once it runs in the right place. Patient data needs a BAA at every layer, and Lovable, Replit and Bolt publish no BAA for hosting today. The code they produce can run in your own cloud account, under your BAA, with the controls HIPAA expects.
Yes. Integrations are core to our work: Epic, athenaOne, eClinicalWorks, labs, payers and X12 clearinghouses. We start vendor approvals in week one. You can map your integrations first in our free planner.
Usually not. In a typical review we keep the screens, flows and much of the data model, and rebuild the parts that hold production back: the backend, auth, data access, tests and infrastructure.
About two weeks of review by our engineers. You get a 14-page report scoring six areas, a week-by-week production plan and a fixed quote. The fee is credited if you build with us, and you keep the report either way.
Most builds take 8 to 12 weeks after the review. The plan in your report sets the real timeline.
In your cloud account, with your keys and your BAA with the provider. We never host your app or your patient data.
Our India team works with sandbox and test data only. Production patient data stays in the US and is handled by your team. Debugging happens in a shared screen session you control. US-based engineers or your own on-site staff are available at extra cost.
Send us the app. We reply within one working day with a start date.