SLED Workshop — Ohio State Student Success Command Center

AI-powered enrollment intelligence demo built on Databricks. Combines a Multi-Agent Supervisor (MAS) with Genie Space for SQL analytics and a Knowledge Assistant for document-based Q&A, all wrapped in a FastAPI + React application.

Architecture

┌─────────────────────────────────────────────────────┐
│  React Frontend (Vite + TailwindCSS)                │
│  ┌──────────┐  ┌──────────┐  ┌────────────────────┐ │
│  │Dashboard │  │ AI Chat  │  │  Architecture View │ │
│  └────┬─────┘  └────┬─────┘  └────────────────────┘ │
├───────┼──────────────┼──────────────────────────────┤
│  FastAPI Backend                                     │
│  /api/metrics ──── SQL Warehouse ──── Gold Tables    │
│  /api/chat ─────── MAS Endpoint ─┬── Genie Space    │
│  /api/conversations              └── Knowledge Asst  │
│       └──────── Lakebase (PostgreSQL)                │
└─────────────────────────────────────────────────────┘

Resource IDs

Resource	ID / Name
App	brickcon-app
App URL	https://brickcon-app-3438839487639471.11.azure.databricksapps.com
SP client_id	5be7b460-a4e7-4703-9144-b813dd38841f
MAS tile	aea8b05a-7532-477c-90f6-4038db05b238
MAS endpoint	mas-aea8b05a-endpoint
KA tile	e236dcf9-8521-4c02-a9ca-3c3e7a24e007
KA endpoint	ka-e236dcf9-endpoint
Genie Space	01f118cd46b51745b597f8b56272b554
Catalog.Schema	demo.sled_workshop
SQL Warehouse	a94a22f8652d85c1
Lakebase Instance	wanderbricks-lakebase
Lakebase PG Schema	sled_workshop
Workspace Profile	dbx_shared_demo

Gold Tables

Table	Description
enrollment_by_college_term	Enrollment counts by college and term with first-gen and regional breakdowns
funnel_by_college_term	Application funnel: applied → admitted → enrolled with admit/yield rates
enrollment_demographics	Enrollment by geographic region, campus, and first-gen status
nursing_pipeline	Nursing program funnel tracking the "double enrollment" initiative

Deployment

Prerequisites

• Databricks CLI configured with profile dbx_shared_demo
• Node.js 18+ (for frontend build)

Deploy

# 1. Build frontend
cd app/frontend && npm install && npm run build && cd ../..

# 2. Deploy bundle (app + job + pipeline definitions)
databricks bundle deploy -t dbx_shared_demo

# 3. Generate data + gold tables (first time or data refresh)
databricks bundle run synthetic_data -t dbx_shared_demo

# 4. Deploy the app
databricks apps deploy brickcon-app --source-code-path app

First-Time Setup (One-Time)

After the first deploy, the app's service principal needs permissions:

-- Unity Catalog
GRANT USE CATALOG ON CATALOG demo TO `5be7b460-a4e7-4703-9144-b813dd38841f`;
GRANT USE SCHEMA ON SCHEMA demo.sled_workshop TO `5be7b460-a4e7-4703-9144-b813dd38841f`;
GRANT SELECT ON SCHEMA demo.sled_workshop TO `5be7b460-a4e7-4703-9144-b813dd38841f`;

Plus API permission grants:

• SQL Warehouse: CAN_USE on a94a22f8652d85c1
• MAS Endpoint: CAN_QUERY on mas-aea8b05a-endpoint
• KA Endpoint: CAN_QUERY on ka-e236dcf9-endpoint
• Genie Space: CAN_RUN on 01f118cd46b51745b597f8b56272b554

Lakebase Setup (One-Time)

Create the sled_workshop PG schema and tables in the wanderbricks-lakebase instance:

CREATE SCHEMA IF NOT EXISTS sled_workshop;
SET search_path TO sled_workshop;

CREATE TABLE conversations (
  id TEXT PRIMARY KEY,
  user_email TEXT NOT NULL,
  title TEXT,
  created_at TIMESTAMPTZ DEFAULT NOW(),
  updated_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE TABLE messages (
  id TEXT PRIMARY KEY,
  conversation_id TEXT NOT NULL REFERENCES conversations(id) ON DELETE CASCADE,
  role TEXT NOT NULL,
  content TEXT NOT NULL,
  route TEXT,
  created_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_messages_conv ON messages(conversation_id, created_at);
CREATE INDEX idx_conversations_user ON conversations(user_email, updated_at DESC);

Project Structure

sled_workshop/
├── databricks.yml              # Asset Bundle config
├── resources/                  # Bundle resource definitions
│   ├── brickcon_app.app.yml    # App resource
│   ├── synthetic_data.job.yml  # Data gen job
│   └── osu_workshop.pipeline.yml  # SDP pipeline (skeleton)
├── app/                        # Databricks App
│   ├── app.py                  # FastAPI entry + middleware
│   ├── app.yaml                # App config (env vars, resources)
│   ├── requirements.txt
│   ├── server/
│   │   ├── config.py           # OAuth + workspace host
│   │   ├── db.py               # Lakebase connection pool
│   │   ├── llm.py              # MAS client (async polling)
│   │   ├── sql_client.py       # SQL Warehouse client
│   │   └── routes/
│   │       ├── chat.py         # /api/chat/* (MAS)
│   │       ├── conversations.py # /api/conversations/* (Lakebase)
│   │       ├── health.py       # /api/health
│   │       └── metrics.py      # /api/metrics/* (SQL)
│   └── frontend/
│       ├── src/
│       │   ├── pages/
│       │   │   ├── Dashboard.tsx
│       │   │   ├── Chat.tsx
│       │   │   └── Architecture.tsx
│       │   └── lib/api.ts
│       └── dist/               # Built frontend (deployed)
├── scripts/
│   └── generate_enrollment_data.py  # Synthetic data + gold tables
└── src/
    └── pipelines/              # SDP pipeline code (skeleton)

Key Design Decisions

• SP token for SQL + MAS: Databricks Apps forward a user token via x-forwarded-access-token, but its scopes are limited to IAM only. All SQL Warehouse and serving endpoint calls use the app's service principal token.
• User token for Lakebase: Lakebase requires the user's forwarded OAuth token (SP tokens lack security labels). The middleware captures it on first request.
• Async polling for MAS: Databricks Apps proxy buffers SSE streams, so we use async task polling (POST to start, GET to poll status) instead of streaming.
• stateMap multi-chat: Chat state is stored per-conversation in a useRef<Map> to prevent animation/message bleed when switching between conversations while a MAS query is in flight.