sanamente.ai

/ The Challenge

Mexican mental-health clinicians run their practices across a patchwork of generic, English-first tools — and lose hours every week to manual clinical documentation that local regulation (NOM-024-SSA3-2012) makes non-negotiable. No Spanish-first, compliance-aware, all-in-one platform existed to unify intake, sessions, clinical notes, treatment plans, scheduling, and patient communication — and none used AI to take the documentation burden off the clinician. So the real challenge wasn't "build an app." It was to prove the opportunity was real, define the right product for how therapists actually work, and get a credible MVP into the market — AI-assisted, and fast — without over-building before the concept was validated, and without cutting the corners that clinical software can't afford to cut.

/ Approach

1. 01

   Found the opportunity in the data, not a hunch. Mined Reddit's practicing-therapist and health-SaaS communities to surface the recurring, still-unsolved pain points of running a therapy practice — documentation load above all — then cross-referenced them against the Mexican market, where localized, compliance-aware, all-in-one tooling is scarce. That produced the thesis: a Spanish-first platform that unifies the entire therapy workflow and uses AI to write the clinical note.

2. 02

   Validated it with the people who'd use it. Ran more than two dozen user interviews with clinicians to pressure-test the pain points and, more importantly, to map how therapists actually work — intake, sessions, notes, treatment plans, scheduling, patient comms. Those conversations, not assumptions, defined the product's core flows and information architecture.

3. 03

   Turned flows into a system and a concept. Translated the validated user flows into a cohesive design system and the initial SanaMente concept: one platform spanning patient charts, a guided session workflow, AI-assisted SOAP notes, treatment plans, a calendar, and patient communication — in Spanish, and shaped around Mexican clinical and privacy standards (NOM-024-SSA3-2012, LFPDPPP).

4. 04

   Prioritized ruthlessly for a rapid MVP. Cut the full concept down to the highest-leverage slice for a credible, fast launch — the session-to-clinical-note core — and put everything else behind feature flags, so the product could ship and start learning without waiting on every module to be perfect.

5. 05

   Now: connecting and testing the whole platform. Re-architected the front end into a backend-ready foundation, wired real authentication, and cleanly separated demo data from live accounts — and is now integrating and testing the platform end to end, ahead of connecting the production database and the AI note-generation pipeline.

/ In the App

/ Selected Technical Challenges

• An abstraction seam engineered to outlive the mock backend. Every data read runs through a single React Query hook layer over typed domain modules, so all call-sites are already async-shaped. Single-record lookups resolve null (never undefined) to satisfy the query cache, and the demo/live switch lives in the query key. The payoff: the entire app can move onto the production database by editing query functions — not by rewriting pages.

• Bugs that were invisible in any single file. Consolidating scattered, duplicated data to one source of truth forced a file-by-file comparison that exposed three real defects no isolated file could reveal: a clinical note whose status disagreed with itself across three sources; a shared component leaking one patient's messages and documents to every patient; and a "mark all as read" that didn't sync between the header dropdown and the notifications page. All three were fixed by unifying each domain onto one cache entry.

• Auth that degrades gracefully — and end-to-end journey testing. Authentication (Clerk) is gated on an environment key, so the product stays fully runnable in a zero-config demo mode and only enforces real accounts when configured — with a route guard and a fully custom Spanish sign-in/up flow (email + 6-digit code) that keeps the brand and clinical tone instead of a generic hosted UI. Tracing the real post-sign-up journey (not just the happy path) surfaced an onboarding flow trap, which was diagnosed and documented for the fix. Every change was verified in a live browser — not just by type-check and build.

/ Outcomes

• Opportunity validated before building — Reddit pain-point mining cross-referenced against the Mexican market, then pressure-tested in more than two dozen user interviews that defined the product's flows and IA.

• A design system and an all-in-one product concept spanning the full therapy workflow, with AI-assisted clinical notes at the center.

• MVP scoped and feature-flagged to the highest-leverage feature set for a rapid, de-risked launch.

• Prototype converted into a secure, backend-ready application — every data read behind one swappable seam, so adopting a real database is a query-function change, not a rewrite.

• Real authentication live end to end — Clerk email/password with 6-digit email-code verification, behind a route guard, with the custom Spanish UI preserved.

• Engineering rigor to match clinical stakes — three latent data-integrity bugs found and fixed, dead code removed, and a phase-scoped, reviewable git history migrated to the production repo with branch protection and backups.

/ Built With