Platforms we have built
Each of these was designed and engineered end to end by our team and is in use today. We've left out product and client names; the descriptions below cover what each system does and how it was built.
Healthcare
Practice management & clinical records
Intake, notes, claims and payments in one system, connected straight to the clearinghouse.
A complete operating system for outpatient therapy practices: intake, scheduling, clinical documentation, treatment plans, insurance claims, family billing, fax, phone and payroll. It was designed around HIPAA from the database up. Every clinic runs in its own database, and the AI only sees the patient information a task requires.
Engineered with
- Next.js
- TypeScript
- PostgreSQL
- AWS ECS
- KMS
- X12 EDI
- SFTP
- Telnyx
- Whisper
- 01 A database per clinicEach practice gets its own Postgres database, with row-level security and a least-privilege application role inside it. One tenant cannot address another's tables.
- 02 Native claims and eligibility837P, 270/271, 276/277, 278, 835 and 999 built in-house and exchanged directly with the clearinghouse over SFTP.
- 03 Denial detectionZero-paid remittances are checked against adjustment-reason codes so denials never post as paid; NCCI bundling conflicts are flagged before submission.
- 04 An AI agent that navigates payer phone treesIt reaches a human and hands the call to a biller. The model asks for a field by name and the application reads out the value, so patient data never reaches the model.
- 05 Dictation that never leaves the deviceSpeech-to-text runs in the clinician's browser. Audio is transcribed locally, not uploaded.
- 06 AI drafting behind a single, contracted chokepointNotes, SMART goals and treatment plans are drafted through one business-associate-covered gateway with no silent fallback to other providers.
- 07 Deny-by-default accessEvery API route must declare who may reach it; a test walks the route tree and fails the build if one doesn't.
- 08 Encryption and backupsKMS encryption at rest, TLS-only storage, point-in-time recovery, cross-region backups and multi-year retention.
Education
School operations & family engagement
Enrollment, billing and parent communication in one product instead of five.
One platform for the schools, academies, camps and programs that usually stitch together five products: enrollment, classes, tuition and billing on one side; class feeds, messaging and family engagement on the other; gradebook, safeguarding, HR and grant-ready impact reporting in between.
Engineered with
- Next.js
- tRPC
- Drizzle
- PostgreSQL RLS
- AWS Fargate
- WAF
- OIDC
- PWA
- 01 Two layers of tenant isolationApplication checks backed by forced row-level security: a connection with no school context sees zero rows.
- 02 Transactional audit logSensitive changes write their audit record inside the same transaction. If the audit fails, so does the change.
- 03 Demographics that can't be traced to a childEquity data for funders is linked by one-way keyed hash, read only in aggregate, and suppressed below a minimum group size.
- 04 Safeguarding with a locked tierIncident and health logs, a restricted safeguarding case file, and live staff-to-child ratio monitoring for licensing.
- 05 Schools keep their own processorFunds settle directly to the school; card data never touches the platform; processor credentials are encrypted per tenant.
- 06 Optional AI draftingTranslation, digests and document extraction run through a swappable model gateway. Output is always a draft, and AI can be switched off per school.
- 07 An MCP server for the school websiteAssistants like Claude can edit a school's site through scoped, single-tenant tokens. Publishing requires a separate permission.
- 08 Multilingual, including right-to-leftEnglish, Spanish and Arabic interfaces with full RTL layout, and compile-time checks for missing translations.
Nonprofit & NGO
Donor CRM & accountable field operations
Tracks each donation from receipt to the field expense it paid for.
A multi-tenant CRM and back office for charities that fund work across borders. It tracks money from the moment a gift arrives, through restricted funds, out to field spending, and back as impact reports. Impact photos carry tamper-evident proof that anyone can check independently.
Engineered with
- Next.js
- tRPC
- Prisma
- PostgreSQL
- Stripe
- Caddy
- OpenTimestamps
- RFC 3161
- Telnyx
- 01 Gift-level traceabilityField spending is matched to the oldest gifts in a fund first, so each donor can see what their gift paid for, without needing an account.
- 02 Restriction-aware overheadAdministrative costs are drawn tier by tier from eligible funds, proportionally and to the cent, honouring donor and religious restrictions.
- 03 Verified captureImpact photos are fingerprinted, written to a hash-chained log, timestamped by an RFC 3161 authority and anchored to Bitcoin hourly.
- 04 Privacy controls on proof pagesProof pages go public only once a report is published; removals for consent or safety redact the image, place and date entirely.
- 05 Custom donor domains, automatic TLSOrganizations point a DNS record; the platform verifies it and issues certificates on demand, with provider-specific setup guidance.
- 06 Donation formsCard and bank payments, fee-coverage nudges, multi-cause baskets, mixed monthly and one-time gifts, and self-service recurring management.
- 07 Texting and voicePer-organization numbers and registrations; opt-outs are re-checked at send time so a STOP holds mid-campaign.
- 08 Built for coding agentsA machine-readable widget registry drives the builder, a public manifest and an MCP server that lets AI tools embed giving forms correctly.
Data & intelligence
Public-data pipelines & scoring systems
Pipelines that combine federal, state and county records and score what they find.
Ingestion, entity resolution and scoring systems built on federal, state and county public data (tax filings, provider registries, civil-rights data and property records), with change detection, quality gates, geospatial matching and privacy suppression built into every run.
Engineered with
- TypeScript
- Python
- PostGIS
- SQLite
- MapLibre
- deck.gl
- Socrata
- ArcGIS REST
- Claude
- 01 Entity resolution across registriesTwo-pass matching links state licensure to the national provider registry: exact identifiers first, then name matching within the same specialty.
- 02 Complete pulls from capped APIsQuery sharding, backoff and resumable ingestion pull complete datasets from endpoints that cap every response.
- 03 Quality-gated releasesContent hashing skips unchanged records; every run is snapshotted and must pass metric gates before it can go live.
- 04 Federal education dataCivil-rights data joined on national district and school identifiers, normalised per thousand students, with small-cell suppression.
- 05 Financial red flags with evidenceTiered findings with severity and cited evidence, and cohort medians and quartiles. Rules compute them and people review them; no language model writes the findings.
- 06 Geospatial matchingParcels match a site only within 75 metres and an agreeing address; weaker evidence is capped and ranked, never assumed.
- 07 County-by-county monitoringAdapters for ArcGIS, Socrata and Carto feeds watch exemptions, liens, tax sales and title transfers across jurisdictions, weekly.
- 08 Language models for classificationSchema-validated, prompt-cached classification replaced a fragile set of rules, with human review before anything is published.
Have a similar problem?
If your team relies on workarounds, manual reconciliation or software that only partly fits, we'd like to hear about it.
Start a conversation →