35 KiB

Raw Blame History

Inou Intake System — Design Specification

Vision

Transform how non-technical users capture health data. No forms. No configuration. Just conversation that crystallizes into structure.

The user talks, the system learns. Yesterday's input becomes today's smart prompt with previous values pre-filled.

Core Principles

Principle	Meaning
Intake AI, not advice AI	We capture, structure, and ask smart follow-ups. We never diagnose or recommend treatment.
Complexity on server, simplicity on client	All intelligence lives server-side. Apps are dumb renderers.
Templates emerge, not configured	User describes leg trainer once → system learns the shape → next time shows fields with previous values.
Conversation, not forms	Onboarding isn't a moment, it's a relationship. Learn one thing at a time, naturally.
Show before save	Always confirm what was understood. Build trust.

Phase 1: Server + Ugly UI

Goal

Fully functional intake system with web-based interface. Prove the model works before investing in pretty UI or mobile.

Deliverables

1. Data Model Extensions

Prompt table:

prompts (
    prompt_id       INTEGER PRIMARY KEY,
    dossier_id      INTEGER NOT NULL,
    category        TEXT,           -- "medication", "vital", "exercise", "symptom"
    type            TEXT,           -- "Lisinopril", "BP", "leg_trainer"
    question        TEXT,           -- "Did you take your morning meds?"
    
    -- Scheduling
    frequency       TEXT,           -- "daily", "weekly:mon,wed,fri", "once", "until_resolved"
    time_of_day     TEXT,           -- "08:00", "12:00", "21:00", NULL for anytime
    next_ask        INTEGER,        -- epoch: next time to show
    expires_at      INTEGER,        -- epoch: stop asking after this (0 = never)
    
    -- Input definition
    input_type      TEXT,           -- "checkbox", "checkbox_group", "number", "text", "scale", "custom"
    input_config    TEXT,           -- JSON: field definitions, scale labels
    
    -- Grouping
    group_name      TEXT,           -- "morning_meds", "gym_session" — for collapsing
    
    -- Source/linking
    trigger_entry   INTEGER,        -- entry that spawned this (for follow-ups)
    created_by      INTEGER,        -- dossier_id of who created (self, caregiver)
    
    -- Last response (for pre-filling)
    last_response       TEXT,       -- JSON: parsed/structured answer
    last_response_raw   TEXT,       -- what they actually said/typed
    last_response_at    INTEGER,    -- epoch
    
    -- State
    dismissed       INTEGER DEFAULT 0,  -- 1 = "don't ask again"
    active          INTEGER DEFAULT 1,  -- 1 = active, 0 = paused
    
    created_at      INTEGER,
    updated_at      INTEGER
)

CREATE INDEX idx_prompts_dossier ON prompts(dossier_id);
CREATE INDEX idx_prompts_dossier_active ON prompts(dossier_id, active, next_ask);

input_config Field Schema

The input_config column stores a JSON object defining form fields. The frontend renders these directly — no AI interpretation needed.

Structure:

{
  "fields": [
    {
      "key": "duration_min",
      "label": "Duration",
      "type": "number",
      "datatype": "int",
      "min": 1,
      "max": 120,
      "unit": "min",
      "required": true
    },
    {
      "key": "intensity",
      "label": "How hard was it?",
      "type": "select",
      "datatype": "string",
      "options": ["easy", "moderate", "hard"],
      "required": true
    },
    {
      "key": "exercises",
      "label": "Exercises",
      "type": "multiselect",
      "datatype": "string[]",
      "options": ["bench", "squat", "deadlift", "treadmill"],
      "required": false
    },
    {
      "key": "notes",
      "label": "Notes",
      "type": "text",
      "datatype": "string",
      "maxlength": 500,
      "required": false
    }
  ]
}

Field types (what to render):

type	datatype	renders as
`text`	string	text input
`textarea`	string	multiline
`number`	int / float	number input with +/-
`checkbox`	bool	single checkbox
`select`	string	dropdown
`multiselect`	string[]	checkboxes or multi-picker
`scale`	int	1-5 or 1-10 buttons
`time`	string	time picker (HH:MM)
`date`	string	date picker

Field properties:

property	applies to	meaning
`key`	all	JSON key in response
`label`	all	display label
`type`	all	widget type
`datatype`	all	validation type
`required`	all	must fill
`default`	all	pre-fill value (overridden by last_response)
`min`	number	minimum value
`max`	number	maximum value
`unit`	number	suffix to display ("°", "min", "lbs")
`step`	number	increment (0.1 for float)
`maxlength`	text	character limit
`options`	select, multiselect	choices
`scale_labels`	scale	e.g. ["terrible", "poor", "ok", "good", "great"]

Response handling:

When rendering a prompt, the frontend:

Reads input_config.fields
For each field, checks last_response[key]
Pre-fills if exists
User submits → stored as new last_response

Entry table additions:

-- Existing entry table, ensure these fields:
entries (
    ...
    parent_id       INTEGER,        -- for nested entries (gym session → exercises)
    data            TEXT,           -- JSON blob for structured details + raw input
    ...
)

Dossier table additions (now live):

-- Added to dossiers table:
    weight_unit     TEXT,           -- "kg" or "lbs"
    height_unit     TEXT,           -- "cm" or "ft"
    last_pull_at    INTEGER,        -- epoch: last time app fetched prompts
    is_provider     INTEGER DEFAULT 0,  -- 1 if healthcare provider
    provider_name   TEXT,           -- practice/clinic name
    away_message    TEXT,           -- auto-responder text
    away_enabled    INTEGER DEFAULT 0   -- 1 if auto-responder active
-- (language already exists on dossier)

2. API Endpoints

GET /api/prompts?dossier=X

Returns prompts due now for this dossier.

Response:

{
  "prompts": [
    {
      "id": "abc123",
      "category": "medication",
      "type": "morning_meds",
      "question": "Did you take your morning medications?",
      "input_type": "checkbox_group",
      "input_config": {
        "items": [
          { "name": "Lisinopril", "amount": "10mg" },
          { "name": "Metoprolol", "amount": "25mg" }
        ]
      },
      "group_name": "morning_meds",
      "group_count": 2,
      "actions": ["took_all", "missed_some", "skip"]
    },
    {
      "id": "def456",
      "category": "exercise",
      "type": "leg_trainer",
      "question": "Leg trainer (morning)",
      "input_type": "custom",
      "input_config": {
        "fields": [
          { "name": "left_min", "label": "Left min", "type": "number", "previous": 20 },
          { "name": "left_angle", "label": "Left °", "type": "number", "previous": 30 },
          { "name": "left_speed", "label": "Left speed", "type": "number", "previous": 4 },
          { "name": "right_min", "label": "Right min", "type": "number", "previous": 20 },
          { "name": "right_angle", "label": "Right °", "type": "number", "previous": 42 },
          { "name": "right_speed", "label": "Right speed", "type": "number", "previous": 5 }
        ],
        "notes_field": true
      },
      "actions": ["save", "skip"]
    },
    {
      "id": "ghi789",
      "category": "followup",
      "type": "symptom_check",
      "question": "Yesterday you had a headache. How are you today?",
      "input_type": "scale",
      "input_config": {
        "options": ["Much better", "Same", "Worse"]
      },
      "actions": ["select", "skip"]
    },
    {
      "id": "open",
      "category": "general",
      "type": "anything_else",
      "question": "Anything else you'd like to mention?",
      "input_type": "freeform",
      "input_config": {
        "placeholder": "Voice, text, or photo...",
        "allow_photo": true,
        "allow_voice": true
      },
      "actions": ["save", "nothing"]
    }
  ],
  "for_dossiers": [
    { "id": "xxx", "name": "Johan", "relation": "self" },
    { "id": "yyy", "name": "Sophia", "relation": "daughter" },
    { "id": "zzz", "name": "Mama", "relation": "mother-in-law" }
  ]
}

POST /api/prompts/respond

Submit response to a prompt.

Request:

{
  "prompt_id": "abc123",
  "dossier_id": "xxx",
  "action": "took_all",
  "response": null
}

Or for custom input:

{
  "prompt_id": "def456",
  "dossier_id": "yyy",
  "action": "save",
  "response": {
    "left_min": 15,
    "left_angle": 25,
    "left_speed": 3,
    "right_min": 15,
    "right_angle": 35,
    "right_speed": 4,
    "notes": "She was tired today"
  }
}

Returns:

{
  "success": true,
  "entries_created": ["entry_id_1", "entry_id_2"],
  "prompt_updated": true
}

POST /api/intake/parse

Parse freeform input (text, voice transcript, photo).

Request:

{
  "dossier_id": "xxx",
  "input_type": "text",
  "content": "Sophia did leg trainer 20 min, left 30 degrees speed 4, right 40 degrees speed 5",
  "attachments": []
}

Or with photo:

{
  "dossier_id": "xxx",
  "input_type": "photo",
  "content": null,
  "attachments": [
    { "type": "image", "data": "base64..." }
  ],
  "context": "supplement_bottle"
}

Returns parsed interpretation for confirmation:

{
  "interpretation": {
    "target_dossier": { "id": "yyy", "name": "Sophia", "confidence": 0.95 },
    "entries": [
      {
        "category": "exercise",
        "type": "leg_trainer",
        "display": "Leg trainer — 20 min",
        "data": {
          "duration_min": 20,
          "left": { "angle": 30, "speed": 4 },
          "right": { "angle": 40, "speed": 5 }
        }
      }
    ],
    "follow_up_question": null,
    "suggested_prompt": {
      "question": "You log leg trainer regularly. Want me to remind you?",
      "options": ["Yes, daily", "Yes, twice daily", "No thanks"]
    }
  },
  "raw": "Sophia did leg trainer 20 min, left 30 degrees speed 4, right 40 degrees speed 5"
}

POST /api/intake/confirm

Confirm parsed interpretation, create entries.

Request:

{
  "interpretation_id": "temp_xxx",
  "confirmed_dossier": "yyy",
  "entries": [...],
  "create_prompt": {
    "answer": "Yes, twice daily"
  }
}

POST /api/intake/ask

Handle intake AI asking follow-up questions.

Request:

{
  "dossier_id": "xxx",
  "context": "unit_clarification",
  "question_type": "weight_unit",
  "user_input": "80"
}

Response:

{
  "question": "Is that kg or lbs?",
  "options": ["kg", "lbs"],
  "store_preference": true
}

GET /api/summary?dossier=X&period=30d

Generate summary for doctor visit.

Response:

{
  "dossier": { "name": "Sophia", "dob": "...", "sex": "female" },
  "period": "last 30 days",
  "sections": {
    "medications": [],
    "therapy": [
      {
        "name": "Leg trainer",
        "frequency": "2x daily",
        "compliance": "12/14 sessions",
        "typical": "L 30° speed 4, R 40° speed 5",
        "trend": "Right side improving (was 35° in December)"
      }
    ],
    "symptoms": [
      { "date": "Dec 28", "type": "headache", "resolution": "resolved after 2 days" }
    ],
    "vitals": {
      "weight": { "start": "32kg", "end": "32.5kg" }
    },
    "imaging": [...],
    "labs": [...],
    "notes": [
      "Tired after school most days",
      "Slept poorly Tuesday, better Wednesday"
    ]
  },
  "markdown": "...",
  "shareable_link": "https://inou.com/s/xxxxx"
}

3. Intake AI Logic (Server-side)

Parsing engine:

Receives text/voice transcript/photo
Identifies target dossier from context ("Sophia", "my son", "I")
Extracts structured data
Detects patterns that could become prompts
Generates follow-up questions when needed

Prompt generation:

From explicit statements: "I take vitamin D every morning" → daily prompt
From repeated behavior: 5 similar entries → suggest prompt
From concerning entries: "running stool" → follow-up tomorrow

Template learning:

First entry with structure → store shape in prompt's input_config
Next time → pre-fill with previous values
Evolves as user input evolves

Multi-dossier routing:

Parse input for person references
Match against accessible dossiers
Require confirmation if ambiguous
Default to last-used dossier if truly unclear

4. Onboarding Flows

Not a wizard. A series of prompts that appear naturally.

Initial (account creation):

Name, DOB, sex — required
That's it. Done.

Day 1-7 (prompted when they return):

"What brings you to inou?" — capture chief concern
"Do you take any daily medications?" — yes → capture list
"Any supplements?" — yes → capture list
"Anything you're tracking? Weight, blood pressure, exercise?"

Week 2+ (contextual):

After first lab upload: "Any conditions we should know about?"
After logging same thing 5x: "Want me to remind you?"
After 2 weeks: "It helps to know family health history. Mind sharing?"

Family history flow:

"Is your mother still living?" → [Yes] [No]
If no: "How old? What happened?" → captures cause
If yes: "Any health conditions?" → open field
Same for father, siblings
"Anything that runs in the family?" → open field

5. Ugly Web UI

Minimal but functional. Prove the model.

Dashboard:

List of prompts due today
Grouped where applicable ("Morning meds (8)")
Each renders based on input_type
"Anything else" always at bottom

Prompt rendering:

checkbox_group: checkboxes + "Took all" button
number: input field with unit
scale: buttons with labels
custom: dynamic fields from input_config
freeform: text area + photo upload + mic button (browser speech-to-text)

Confirmation modal:

Shows parsed interpretation
Allows dossier change if wrong
Edit button for corrections
Save / Cancel

History view:

Today's entries
Past 7 days
Searchable
Editable (tap to modify)

Summary view:

Generate summary for selected period
Copy as text
Print
Share link

Phase 2: Pretty UI + Mobile

Goal

Native mobile experience with Flutter. Beautiful UI. Local capabilities (speech-to-text, barcode, camera).

Deliverables

1. Flutter App

Architecture:

Calls same API endpoints
Local speech-to-text (send transcript to server)
Local barcode scanning (send code to server for lookup)
Local OCR first pass (send text + image if low confidence)
Local notifications scheduled from prompt data

Pull model:

On app open: GET /api/prompts
Background refresh: once daily
Store prompts locally for offline
Queue responses if offline, sync when connected

Push fallback:

Server tracks last_pull_at per dossier
If no pull in 3+ days AND prompts due → push notification
Push just says "Don't forget to log" — opens app → pull → render

Screens:

Home (Prompts):

Cards for each due prompt
Swipe to skip
Tap to expand/respond
Grouped cards for multi-item (morning meds)
Always: "Anything else?" at bottom

Input:

Big mic button for voice
Camera for photos (bottle labels, wounds, readings)
Text field
Confirmation before save

Log (History):

Timeline view
Filter by category, dossier
Tap to edit

Summary:

Pick dossier, pick period
Generate and display
Share / Export

Dossiers:

Switch between accessible dossiers
Default for logging (remembers last used)

2. Notification System

Local notifications:

App schedules based on prompt times
"Morning meds" at 8:00
"Leg trainer (evening)" at 18:00
Tapping opens app directly to that prompt

Server push:

Firebase Cloud Messaging
Only when app hasn't pulled in days
Generic "Check in with inou"

3. Offline Support

Cached data:

Prompts (with input_config, previous values)
Recent entries
Dossier list

Queued responses:

If offline, store locally
Sync on reconnect
Show "pending" indicator

4. Voice Flow

User taps mic:

Device speech-to-text (on-device)
Transcript sent to server: POST /api/intake/parse
Server parses, returns interpretation
App shows confirmation
User confirms → entry created

Natural language examples:

"Sophia did leg trainer, 20 minutes, left 30 degrees speed 4, right 40 degrees speed 5"
"I took 2 Tylenol for a headache"
"Mom took her pills"
"Blood pressure 120 over 80"
"Walked for an hour this morning"

5. Photo Flow

Supplement bottle:

User taps camera on "Any supplements?" prompt
Takes photo of bottle
Device OCR extracts text (or barcode scan)
Server: looks up product, extracts dose
Returns: "Vitamin D3 5000 IU — when do you take this?"
User selects [Morning] [Evening] etc.
Prompt created

Vitals (BP monitor, scale):

Photo of display
OCR extracts numbers
Confirm values
Entry created

Injury/symptom:

Photo of knee scrape
User adds context: "My son fell"
Entry created with photo attached

6. Pretty UI Design

Follow existing inou aesthetic:

Warm, not clinical
IBM Plex Sans
Desert palette: warm beige (#F5EDE4), dark brown (#4A3728), terracotta (#C4704B)

Cards with generous padding. Large tap targets. Clear typography.

Accessible: works for someone with shaky hands or poor vision.

Appendix A: Example Prompt Types

Type	Input	Previous values?	Follow-up?
Medication (single)	checkbox	N/A	If missed 3+ days
Medications (group)	checkbox_group + "took all"	N/A	If missed
Vital (number)	number field + unit	Optional	If concerning
Supplement	checkbox	N/A	No
Exercise (simple)	number (duration)	Yes	If goal missed
Exercise (complex)	custom fields	Yes, all fields	No
Symptom follow-up	scale	N/A	If "worse"
Freeform	text + voice + photo	N/A	Based on content

Appendix B: Entry Structure Examples

Simple medication:

{
  "category": "medication",
  "type": "Lisinopril",
  "value": "taken",
  "data": { "dose": "10mg" },
  "timestamp": 1704369600
}

Leg trainer session:

{
  "category": "exercise",
  "type": "leg_trainer",
  "value": "20 min",
  "data": {
    "duration_min": 20,
    "left": { "angle": 30, "speed": 4 },
    "right": { "angle": 40, "speed": 5 },
    "notes": "She was tired today",
    "raw": "Sophia did leg trainer, 20 min..."
  },
  "timestamp": 1704369600
}

Gym session with nested entries:

// Parent
{
  "id": "gym_001",
  "category": "exercise",
  "type": "gym_session",
  "value": null,
  "timestamp": 1704369600
}

// Children
{
  "parent_id": "gym_001",
  "category": "exercise",
  "type": "bench_press",
  "value": "200",
  "data": { "unit": "lbs" }
}
{
  "parent_id": "gym_001",
  "category": "exercise",
  "type": "squats",
  "value": "150",
  "data": { "unit": "lbs" }
}

Supplement from photo:

{
  "category": "supplement",
  "type": "Vitamin D3",
  "value": "1",
  "data": {
    "dose": "5000 IU",
    "brand": "Nature's Way",
    "source": "photo_ocr",
    "photo_id": "attach_xxx"
  },
  "timestamp": 1704369600
}

Family history:

{
  "category": "family_history",
  "type": "mother",
  "value": "deceased",
  "data": {
    "age_at_death": 78,
    "cause": "heart attack",
    "conditions": ["diabetes", "hypertension"],
    "raw": "She had diabetes and high blood pressure, died of a heart attack at 78"
  }
}

Appendix C: Intake AI Prompt Guidelines

Role: You are an intake assistant for a health tracking system. Your job is to:

Parse natural language into structured health data
Ask clarifying questions when needed
Identify which person (dossier) an entry is about
Never give medical advice or interpret results

Parsing rules:

Extract: category, type, value, relevant details
Preserve raw input in data.raw
Resolve aliases (D3 → Vitamin D3, BP → blood pressure)
Detect time references ("yesterday", "this morning")
Detect person references ("my son", "Sophia", "mom", "I")

Multi-dossier:

If person mentioned matches an accessible dossier → route there
If ambiguous ("blood pressure 120/80") → ask who
If clearly about self ("I walked") → current user's dossier

Follow-up questions:

Unknown unit: "Is that kg or lbs?"
New supplement: "When do you take this?"
Symptom logged: create follow-up prompt for tomorrow
Repeated activity: "Want me to remind you about this?"

What NOT to do:

Don't interpret if values are good/bad
Don't recommend treatments
Don't diagnose
Don't say "you should see a doctor" (unless asked about resources)

Appendix D: Implementation Order

Phase 1 — Server + Ugly UI

Database: Add prompts table, extend entries
API: /api/prompts, /api/prompts/respond
Web UI: Render prompts, handle responses
Intake parsing: POST /api/intake/parse (start simple, text only)
Confirmation flow: Show interpretation, confirm, save
Prompt generation: From explicit statements ("I take X daily")
Follow-up logic: Symptoms trigger next-day prompts
Template learning: Remember structure from first entry
Previous values: Pre-fill from last entry of same type
Grouping: Collapse multiple meds into "took all"
Multi-dossier: Parse who, confirm, route
Onboarding prompts: Medications, supplements, goals
Summary generation: /api/summary
Family history flow: Guided questions

Phase 2 — Pretty UI + Mobile

Flutter scaffold: Navigation, screens
API integration: Same endpoints
Prompt rendering: All input types
Voice input: On-device STT → server parse
Photo input: Camera, OCR, send to server
Barcode: On-device scan → server lookup
Local notifications: Schedule from prompt times
Offline support: Cache prompts, queue responses
Push fallback: Firebase integration
Polish: Design system, animations, accessibility

Document version: 1.0 Last updated: January 5, 2025

Phase 1b: Provider Chat

Goal

Secure messaging between patients/caregivers and healthcare providers. Replace Spruce for basic communication. FIPS 140-3 encrypted.

Why Providers Switch

Spruce	Inou
$24-49/user/month	Free
Just messages	Messages + patient's complete dossier
HIPAA	HIPAA + FIPS 140-3

Data Model (now live)

Messages table:

messages (
    message_id      INTEGER PRIMARY KEY,
    conversation_id INTEGER NOT NULL,
    dossier_id      INTEGER NOT NULL,     -- which patient dossier this relates to
    sender_id       INTEGER NOT NULL,
    recipient_id    INTEGER NOT NULL,
    
    body            TEXT,                 -- encrypted
    attachments     TEXT,                 -- encrypted JSON
    
    sent_at         INTEGER NOT NULL,
    read_at         INTEGER,
    
    auto_reply      INTEGER DEFAULT 0
)

CREATE INDEX idx_messages_conversation ON messages(conversation_id, sent_at);
CREATE INDEX idx_messages_recipient_unread ON messages(recipient_id, read_at);

Conversations table:

conversations (
    conversation_id INTEGER PRIMARY KEY,
    dossier_id      INTEGER NOT NULL,     -- the patient dossier
    provider_id     INTEGER NOT NULL,     -- the provider's dossier
    
    last_message_at INTEGER,
    unread_count    INTEGER DEFAULT 0,
    
    created_at      INTEGER
)

CREATE INDEX idx_conversations_dossier ON conversations(dossier_id);
CREATE INDEX idx_conversations_provider ON conversations(provider_id);

Provider fields (added to dossiers table):

-- See Phase 1 dossier additions for:
--   is_provider, provider_name, away_message, away_enabled

API Endpoints

GET /api/conversations?dossier=X

List all conversations for a dossier (as patient or provider).

{
  "conversations": [
    {
      "id": "conv_123",
      "patient": { "id": "xxx", "name": "Sophia" },
      "provider": { "id": "yyy", "name": "Dr. Patel", "practice": "Kids First Pediatrics" },
      "last_message": {
        "body": "See you at the appointment tomorrow",
        "sent_at": 1704369600,
        "sender": "provider"
      },
      "unread_count": 0
    }
  ]
}

GET /api/messages?conversation=X&limit=50&before=timestamp

Get messages in a conversation (paginated).

{
  "messages": [
    {
      "id": "msg_456",
      "sender": { "id": "xxx", "name": "Johan", "role": "caregiver" },
      "body": "Sophia's leg trainer went well today. See attached notes.",
      "attachments": [
        { "type": "entry", "id": "entry_789", "name": "Leg trainer session" }
      ],
      "sent_at": 1704369600,
      "read_at": 1704369700
    }
  ]
}

POST /api/messages

Send a message.

{
  "conversation_id": "conv_123",
  "body": "Quick question about her medication dosage",
  "attachments": []
}

Response:

{
  "id": "msg_789",
  "sent_at": 1704369600,
  "auto_reply": {
    "id": "msg_790",
    "body": "Thank you for your message. I'm currently unavailable but will respond within 24 hours. For urgent matters, please call 555-1234.",
    "sent_at": 1704369601
  }
}

POST /api/messages/:id/read

Mark message as read.

PUT /api/provider/settings

Update provider settings (away message, etc).

{
  "away_enabled": true,
  "away_message": "Thank you for your message. I typically respond within 24 hours. For urgent matters, call 555-1234."
}

Features

Core:

Text messages (encrypted at rest, FIPS)
Attach entries from dossier (labs, imaging, notes)
Attach photos
Read receipts
Auto-responder when provider unavailable

Notifications:

Push notification on new message
Email notification if app not opened in 24h

Provider View:

List of patient conversations
Unread count badge
Click patient → see conversation + dossier summary
Quick access to recent entries, labs, imaging

Provider Onboarding

Invited by patient:

Patient sends invite to provider email
Provider receives: "Johan invited you to view Sophia's health dossier"
Provider signs up (free): name, email, practice name
Provider linked to patient's dossier with role = medical
Provider can message, view dossier

Provider invites patients:

Provider creates account (free)
Provider uploads patient list (email CSV) or invites one by one
Patient receives: "Dr. Patel invited you to connect on inou"
Patient signs up → linked to provider
Patient can message, optionally upgrade to full tracking

Patient Tiers

Tier	Cost	Features
Passive	Free	Receive messages, basic profile, view shared docs
Active	Paid	Full dossier: intake AI, tracking, prompts, summaries, imaging, labs, genome

Provider brings patients in free. Some upgrade. That's the revenue.

Auto-Responder Logic

func sendMessage(msg Message) {
    save(msg)
    notify(recipient)
    
    if recipient.is_provider && recipient.away_enabled {
        autoReply := Message{
            sender: recipient,
            recipient: msg.sender,
            body: recipient.away_message,
            auto_reply: true,
        }
        save(autoReply)
        notify(msg.sender)
    }
}

Security

All messages encrypted with FIPS 140-3 (AES-256-GCM)
TLS 1.3 in transit
Messages only visible to conversation participants
Providers can only see dossiers they're invited to
Audit log for all access

Implementation Order

Database: messages, conversations tables; provider fields on dossier
API: conversations, messages, send, read
Web UI: conversation list, message thread, compose
Auto-responder: provider settings, auto-reply logic
Notifications: push (Firebase), email fallback
Provider invite flow: patient invites provider
Bulk invite: provider imports patient list
Provider dashboard: patient list, unread counts, dossier access

Document version: 1.1 Last updated: January 5, 2025

Appendix E: Entry Categories

Complete list of entry categories supported by the system.

Core (Daily Logging)

Category	Description	Example Types
`medication`	Medications taken or prescribed	Lisinopril, Metoprolol, anti-seizure
`vital`	Vital signs and measurements	BP, weight, temperature, SpO2
`exercise`	Physical activity and therapy sessions	leg_trainer, walking, gym_session
`symptom`	Symptoms and complaints	headache, seizure, fatigue
`supplement`	Supplements and vitamins	Vitamin D3, magnesium, iron
`note`	Free-form notes	daily observations

Events

Category	Description	Example Types	Nesting
`surgery`	Surgical procedures	shunt_placement, shunt_revision, ETV	Parent with child `procedure` entries
`hospitalization`	Hospital stays	NICU, inpatient, ER	Parent with child entries for daily events
`consultation`	Doctor visits and consults	video_call, office_visit, second_opinion	Parent with child `finding` or `recommendation` entries

Conditions

Category	Description	Example Types	Notes
`diagnosis`	Medical diagnoses	IVH, hydrocephalus, epilepsy, ROP	Can have status: active, resolved, suspected
`device`	Implanted or external devices	VP_shunt, Ommaya_reservoir, cochlear_implant	Track settings changes over time as new entries

Development

Category	Description	Example Types	Notes
`therapy`	Therapeutic interventions	Vojta, Bobath, Feldenkrais, ABM, speech_therapy	Include outcome: effective, ineffective, ongoing
`assessment`	Developmental evaluations	motor, speech, cognitive, vision	Qualitative findings, milestones

Foundational

Category	Description	Example Types	Notes
`birth`	Birth and prenatal history	delivery, Apgar, gestational_age, complications	One-time entries, foundational context
`history`	Past medical events without full documentation	surgery, illness, injury, allergy	Brief recollections: "appendectomy at age 12"
`family_history`	Family medical history	mother, father, sibling, maternal_grandmother	Include condition, age at diagnosis, outcome

Findings (Interpretations)

Category	Description	Example Types	Notes
`imaging_finding`	Radiologist or AI interpretation	MRI_finding, CT_finding, XR_finding	Links to DICOM study via data.study_guid
`eeg_finding`	EEG report findings	epileptiform_activity, sleep_pattern	From EEG monitoring reports

Category	Description	Example Types	Notes
`provider`	Healthcare providers	neurosurgeon, neurologist, therapist	Institution, contact, relationship period
`question`	Open medical questions	surgical, diagnostic, treatment	Answer via child entry when resolved
`upload`	Uploaded files	imaging, genetics, document, lab_report	Existing category

Nesting Examples

Complex surgery with sub-procedures:

Parent:
  category: surgery
  type: shunt_revision
  value: "Complex Shunt Revision"
  timestamp: 2021-12-10
  data: {
    location: "University Clinic Ulm",
    surgeon: "Dr. Aurelia Peraud",
    indication: "Isolated 4th ventricle with brainstem compression"
  }

Children (parent_id → parent):
  category: procedure, type: shunt_replacement, data: {old: "Codman", new: "Miethke Blu"}
  category: procedure, type: catheter_placement, data: {target: "4th ventricle"}
  category: procedure, type: ETV, data: {location: "3rd ventricle floor"}
  category: procedure, type: catheter_placement, data: {from: "3rd ventricle", to: "anterior brain stem"}

Device with settings history:

-- Initial installation --
category: device
type: miethke_blu_valve
value: "installed"
timestamp: 2021-12-10
data: {setting: "20 cmH2O", component: "M. blue plus", surgeon: "Dr. Peraud"}

-- Setting adjustment --
category: device
type: miethke_blu_valve
value: "adjusted"
timestamp: 2022-08-15
data: {setting: "30 cmH2O", adjusted_by: "Ulm team"}

-- Suspected malfunction --
category: device
type: miethke_blu_valve
value: "malfunction"
timestamp: 2024-10-21
data: {expected_setting: "30 cmH2O", actual_setting: "24 cmH2O", status: "suspected dysfunction"}

Open question with answer:

Parent:
  category: question
  type: surgical
  value: "Is there opportunity for endoscopy to rearrange natural CSF flow?"
  timestamp: 2025-01-01
  data: {status: "unanswered", raised_by: "parent"}

Child (when answered):
  parent_id: [question entry]
  category: answer
  type: specialist_opinion
  value: "Dr. Peraud recommends shunt revision over endoscopy due to..."
  timestamp: 2025-02-15
  data: {source: "video consultation", provider: "Dr. Peraud"}

Consultation with findings:

Parent:
  category: consultation
  type: video_call
  value: "Ulm Neurosurgery Follow-up"
  timestamp: 2024-10-21
  data: {
    institution: "University Clinic Ulm",
    provider: "Prof. Dr. Aurelia Peraud",
    interpreter: "Frau Natalya Kazmina"
  }

Children:
  category: finding, type: imaging, value: "Ventricular enlargement progression"
  category: finding, type: device, value: "Suspected valve dysfunction"
  category: recommendation, type: surgical, value: "Shunt and valve revision recommended"
  category: recommendation, type: screening, value: "MRSA/MRGN testing required pre-op"

Personal medical history (undocumented past events):

category: history
type: surgery
value: "Appendectomy"
data: {age: 12, approximate_year: 2005, notes: "no complications"}

category: history
type: illness
value: "Chickenpox"
data: {age: 6, severity: "normal"}

category: history
type: injury
value: "Broke left arm"
data: {year: 2015, treatment: "cast 6 weeks"}

category: history
type: allergy
value: "Penicillin"
data: {reaction: "rash", severity: "moderate", discovered: "childhood"}

Family history:

category: family_history
type: mother
value: "Breast cancer"
data: {age_at_diagnosis: 58, outcome: "survived", treatment: "mastectomy + chemo"}

category: family_history
type: father
value: "Type 2 diabetes"
data: {age_at_diagnosis: 52, status: "ongoing", management: "metformin"}

category: family_history
type: maternal_grandmother
value: "Alzheimer's disease"
data: {age_at_onset: 72, age_at_death: 81}

category: family_history
type: brother
value: "Healthy"
data: {age: 35, notes: "no known conditions"}

category: family_history
type: family_pattern
value: "Heart disease"
data: {notes: "Multiple paternal relatives with early heart attacks"}

35 KiB Raw Blame History

Inou Intake System — Design Specification

Vision

Core Principles

Phase 1: Server + Ugly UI

Goal

Deliverables

1. Data Model Extensions

input_config Field Schema

2. API Endpoints

3. Intake AI Logic (Server-side)

4. Onboarding Flows

5. Ugly Web UI

Phase 2: Pretty UI + Mobile

Goal

Deliverables

1. Flutter App

2. Notification System

3. Offline Support

4. Voice Flow

5. Photo Flow

6. Pretty UI Design

Appendix A: Example Prompt Types

Appendix B: Entry Structure Examples

Appendix C: Intake AI Prompt Guidelines

Appendix D: Implementation Order

Phase 1 — Server + Ugly UI

Phase 2 — Pretty UI + Mobile

Phase 1b: Provider Chat

Goal

Why Providers Switch

Data Model (now live)

API Endpoints

Features

Provider Onboarding

Patient Tiers

Auto-Responder Logic

Security

Implementation Order

Appendix E: Entry Categories

Core (Daily Logging)

Events

Conditions

Development

Foundational

Findings (Interpretations)

Meta

Nesting Examples

35 KiB

Raw Blame History