File size: 9,549 Bytes
dcdaf35 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 | ---
title: AI Messaging System
emoji: 🎶
colorFrom: blue
colorTo: gray
sdk: streamlit
sdk_version: 1.50.0
python_version: 3.9
app_file: app.py
---
# AI Messaging System v3 — Visualization Tool
A Streamlit tool for reviewing and evaluating AI-generated push notification messages. All data is read from a single output table — no joins. Reviewers filter messages by brand, campaign, and scenario, inspect user context stored at generation time, and flag any message that has issues. Rejections are persisted to Snowflake. No action on a message means it is good.
This tool does **not** generate messages. It is purely for reviewing messages already produced by the v3 pipeline.
---
## Purpose
- Pre-fetch the most recent 150 messages per campaign on login (single query, no joins)
- Filter instantly by brand, campaign, and scenario — all client-side, no extra DB queries
- View the user context captured at generation time (streaks, profile, previous messages, last interacted content)
- Reject messages with a structured reason — persisted to Snowflake immediately
- Track rejection counts per review session in the sidebar
---
## Directory Structure
```
visualization/
├── app.py # Entry point: login + data pre-fetch on authentication
├── pages/
│ └── 1_Message_Viewer.py # Message cards, filters, user context, rejection UI
├── utils/
│ ├── auth.py # Login / session authentication
│ ├── snowflake_client.py # Snowflake reads (single table) and feedback writes
│ ├── feedback_manager.py # In-session rejection state + Snowflake persistence
│ ├── theme.py # Brand colours, campaign labels, scenario definitions
│ └── __init__.py
├── requirements.txt
├── .env # Credentials (not committed)
├── .env.example
└── README.md
```
---
## Getting Started
### 1. Install dependencies
```bash
pip install -r requirements.txt
```
### 2. Configure credentials
Copy `.env.example` to `.env` and fill in your Snowflake credentials and the app token:
```
SNOWFLAKE_USER = ...
SNOWFLAKE_PASSWORD = ...
SNOWFLAKE_ACCOUNT = ...
SNOWFLAKE_ROLE = ACCOUNTADMIN
SNOWFLAKE_DATABASE = RECSYS_V3
SNOWFLAKE_WAREHOUSE= COMPUTE_WH
SNOWFLAKE_SCHEMA = PUBLIC
APP_TOKEN = ...
```
### 3. Run the app
```bash
cd ai_messaging_system_v3/visualization
streamlit run app.py
```
---
## Authentication
Login requires an email from the authorised list and a shared access token (`APP_TOKEN` in `.env`). Authorised emails are defined in `utils/auth.py`.
---
## Snowflake Tables
### Input (read-only) — single table, no joins
All data needed for display and filtering is stored at message generation time in one table.
**`MESSAGING_SYSTEM_V2.GENERATED_DATA.DAILY_PUSH_MESSAGES`**
| Column | Type | Description |
| --- | --- | --- |
| `USER_ID` | NUMBER | Target user |
| `DETECTED_BRAND` | VARCHAR | Brand: `drumeo`, `pianote`, `guitareo`, `singeo`, `playbass` |
| `BRANCH` | VARCHAR | Campaign routing key (mirrors `CAMPAIGN_NAME`) |
| `MESSAGE` | VARCHAR | JSON string `{"header": "...", "message": "..."}` — parsed into header + body |
| `PLATFORM` | VARCHAR | Delivery platform (e.g. `push`) |
| `CAMPAIGN_NAME` | VARCHAR | `dailyPush_dailyStreak`, `dailyPush_weeklyStreak`, `dailyPush_noStreak` |
| `TIMESTAMP` | TIMESTAMP | Generation timestamp |
| `RECOMMENDATION` | VARCHAR | Recommendation type; `for_you` means no specific recommendation (used for scenario detection) |
| `RECOMMENDED_CONTENT_ID` | NUMBER | Content ID linked in the message (if any) |
| `FIRST_NAME` | VARCHAR | User's first name at generation time |
| `CURRENT_DAILY_STREAK_LENGTH` | NUMBER | Daily streak at generation time |
| `CURRENT_WEEKLY_STREAK_LENGTH` | NUMBER | Weekly streak at generation time |
| `USER_PROFILE` | VARCHAR | User profile text used for personalisation |
| `PREVIOUS_MESSAGES` | VARCHAR | JSON snapshot of prior messages for this user |
| `LAST_INTERACTED_CONTENT_PROFILE` | VARCHAR | Profile of the last content the user interacted with |
### Output (written by this tool)
**`MESSAGING_SYSTEM_V2.UI.V3_FEEDBACKS`** — created automatically on first run.
| Column | Description |
| --- | --- |
| `BATCH_ID` | UUID generated per review session |
| `USER_ID` | Reviewed user |
| `CAMPAIGN_NAME` | Campaign of the reviewed message |
| `BRAND` | Brand of the reviewed user |
| `REJECTION_REASON` | Category key (see Rejection Reasons) |
| `REJECTION_TEXT` | Optional free-text note from the reviewer |
| `MESSAGE_HEADER` | Full header at time of review |
| `MESSAGE_BODY` | Full body at time of review |
| `MESSAGE_TIMESTAMP` | When the message was originally generated |
| `REVIEWER_EMAIL` | Email of the reviewer |
| `TIMESTAMP` | When the rejection was submitted |
Rejections are upserted via `MERGE` — changing the reason on an already-rejected message updates the existing row.
Users with any existing rejection in `V3_FEEDBACKS` are **excluded** from the pre-fetch query, so the same message is never shown twice across sessions.
---
## Filters
All filters are applied instantly client-side on the pre-loaded DataFrame — no Snowflake query is triggered on filter change.
| Filter | Source column | Notes |
| --- | --- | --- |
| Brand | `DETECTED_BRAND` | All, Drumeo, Pianote, Guitareo, Singeo, Playbass |
| Campaign | `CAMPAIGN_NAME` | All, Daily Streak, Weekly Streak, No Streak |
| Scenario | `RECOMMENDATION` | Weekly Streak campaign only (see Scenarios) |
| Only users with previous messages | `HAS_PREVIOUS` | Pre-computed at fetch time via `COUNT(*) > 1` |
---
## Campaigns and Scenarios
| Campaign | Scenario key | Display label | Detection |
| --- | --- | --- | --- |
| `dailyPush_dailyStreak` | — | — | No sub-scenarios |
| `dailyPush_weeklyStreak` | `no_practice_this_week` | No Practice This Week | `RECOMMENDATION == 'for_you'` |
| `dailyPush_weeklyStreak` | `practiced_this_week` | Practiced This Week | `RECOMMENDATION != 'for_you'` |
| `dailyPush_noStreak` | — | — | No sub-scenarios |
---
## Rejection Reason Categories
| Key | Label |
| --- | --- |
| `poor_header` | Poor Header |
| `poor_body` | Poor Body / Content |
| `grammar_issues` | Grammar Issues |
| `emoji_problems` | Emoji Problems |
| `recommendation_issues` | Recommendation Issues |
| `wrong_information` | Wrong / Inaccurate Information |
| `tone_issues` | Tone Issues |
| `similarity` | Similar To Previous Messages |
| `other` | Other |
---
## Supported Brands
| Brand | Colour |
| --- | --- |
| Drumeo | Red `#E84545` |
| Pianote | Blue `#4A90D9` |
| Guitareo | Amber `#F5A623` |
| Singeo | Purple `#7B68EE` |
| Playbass | Green `#3DAA5C` |
---
## UI Flow
```
app.py (Login form)
↓ authenticated
app.py (Landing)
→ pre-fetches top 150 messages per campaign from DAILY_PUSH_MESSAGES
→ stores result in session_state["all_messages"]
→ shows per-campaign counts
↓ navigate to Message Viewer
pages/1_Message_Viewer.py
Sidebar (instant, no Apply button):
Brand / Campaign / Scenario / "Only users with prev messages"
Reload button → clears cache, re-fetches
Sidebar:
Session stats: showing · rejected · reject %
Main area — paginated cards (20 per page):
Each card:
┌─ Brand badge · Campaign · User #ID · Generated timestamp ──┐
│ │
│ Push Notification Preview │
│ [Header text] │
│ [Body text] │
│ Header: N chars · Body: N chars │
│ │
│ ▼ User Context (collapsible) │
│ First name · Daily streak · Weekly streak · Profile │
│ │
│ ▼ Previous Messages (collapsible) │
│ Parsed from PREVIOUS_MESSAGES column — no extra query │
│ │
│ ▼ Last Interacted Content Profile (collapsible) │
│ │
│ Feedback column: │
│ [❌ Reject] → reason dropdown + optional note → Submit │
│ Already rejected: shows reason · [✏️ Change] [Clear] │
└─────────────────────────────────────────────────────────────┘
→ Rejections saved to Snowflake on Submit
→ No action = message is considered good
```
---
## Module Responsibilities
| Module | Responsibility |
| --- | --- |
| `auth.py` | Session authentication, email allowlist, token verification |
| `snowflake_client.py` | Single-table fetch from `DAILY_PUSH_MESSAGES`; feedback MERGE/DELETE |
| `feedback_manager.py` | In-session rejection dict (fast cache), delegates writes to `snowflake_client` |
| `theme.py` | Brand colours, campaign labels, scenario definitions and `detect_scenario()` | |