🏆 Devpost: FinHack

AI finance agents that live in your real-life context — capture purchases, learn your habits, find better deals, and coach you into budget-aware, shared saving… Women-first.

FinHack is a women-first personal finance experience + an internal ops dashboard.

• For users: low-friction spending capture, budget-aware guidance, and smarter purchase discovery.
• For teams: a Company View dashboard that monitors trends, anomalies, and system health — with an Employee Assistant that explains what’s happening in plain language.

If budgeting apps feel like homework, this is the opposite: supportive, contextual, and designed to be used every day.

Many people disengage from finance tools because:

• Tracking is tedious (high friction).
• Feedback feels judgmental instead of supportive.
• Most tools assume stable income + solo decision-making.
• Shared saving (roommates/partners/friends) lacks clarity and accountability.

We built Finance Queens to make saving feel encouraging, practical, and repeatable.

Capture purchases naturally (speak instead of type). We extract key details (amount/category/note) and update insights in real-time.

A smart “better deal” layer that compares options and can prioritize women-owned / women-led choices (optional focus mode).

The app learns patterns and nudges users with actionable next steps (not guilt). Think: “here’s one small move today.”

We use fast forecasting on transaction aggregates to predict spend and flag potential overspend risk for the week ahead.

A Streamlit dashboard for the team: users, transactions, alerts, trends, and AI assistants to speed up investigation and support.

Note: Some integrations are “prototype/optional” for the hackathon demo, but the architecture is built to support them cleanly.

User (web/mobile)
   │
   ▼
Node/Express API (apps/llm)  ──► External AI/Tools (Gemini, ElevenLabs, SerpAPI)
   │
   ├─► Data layer (Supabase / Snowflake)
   │
   └─► Company View (Streamlit dashboard) (apps/company_view)
           └─► Employee Assistant + Support Assistant

Deployment: Docker → Vultr VM (backend + dashboard)
Optional: Solana Devnet audit log (risk flags / receipts / events)

cd apps/llm
npm install
PORT=8080 npm run start

# health check
curl -sS http://127.0.0.1:8080/v1/health

cd apps/company_view
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

streamlit run app.py --server.port 8501

Open: http://localhost:8501

Create apps/company_view/.env (dotenv format, KEY=value):

# backend base
LLM_BASE_URL=http://127.0.0.1:8080

# optional live data (dashboard falls back to demo/mock data if not set)
SUPABASE_URL=
SUPABASE_SERVICE_ROLE_KEY=

# assistants
GEMINI_API_KEY=
GEMINI_MODEL=gemini-2.5-flash

# optional
REQUEST_TIMEOUT_S=30

We use Snowflake ML Forecast to generate a 7‑day spend prediction + uncertainty band (fast enough for hackathon demos):

CREATE OR REPLACE SNOWFLAKE.ML.FORECAST SPEND_FORECAST_DEMO (
  INPUT_DATA => SYSTEM$REFERENCE('TABLE', 'FINHACK_DB.ANALYTICS.DAILY_SPEND_DEMO'),
  SERIES_COLNAME => 'USER_ID',
  TIMESTAMP_COLNAME => 'DAY',
  TARGET_COLNAME => 'SPEND',
  CONFIG_OBJECT => {
    'frequency': '1 day',
    'method': 'fast',
    'evaluate': false
  }
);

SELECT *
FROM TABLE(SPEND_FORECAST_DEMO!FORECAST(FORECASTING_PERIODS => 7));

We containerize the backend with Docker and deploy to a Vultr VM with a public health endpoint for judges to verify uptime.

Typical flow:

# build (from repo root)
docker build -t finhack-llm:latest -f apps/llm/Dockerfile .

# run
docker rm -f finhack-llm 2>/dev/null || true
docker run -d --name finhack-llm --restart unless-stopped -p 8080:8080 finhack-llm:latest

# verify
curl -sS http://127.0.0.1:8080/v1/health

Finance Queens is built for the real world:

• Less friction → more consistency. Voice capture makes tracking sustainable.
• Supportive coaching. We replace guilt-based budgeting with small, actionable next steps.
• Shared saving clarity. Transparent goals make group saving less awkward.
• Earlier risk signals. Forecasting + anomaly detection helps users (and teams) act before a problem compounds.

apps/
  llm/            # Node/Express API + AI endpoints
  company_view/   # Streamlit internal dashboard + assistants

MIT