Finlo (finance + flow)

architecture diagram
landing page
fastapi backend
database - supabase

Inspiration

Small businesses juggle receipts, spreadsheets, and pricey accountants just to know if they can make payroll. We wanted a calmer “flow” for SMB's (Small-Medium Sized Businesses), a single brain that watches the money 24/7 and whispers what to do next.

What it does

Finlo is an AI CFO + finance OS. Drop an expense in plain English (“paid rent $2 000”), and Finlo instantly:

Posts it to Supabase with the right negative / positive sign
Flags anomalies and category trends with live rules
Forecasts revenue, expenses, and cash-flow 30–90 days ahead using Prophet
Returns clear insights (“profit margin trending ↓ 8%”) that owners can act on.

How we built it

FastAPI backend exposes REST endpoints
Groq Llama 3 turns natural-language requests into SQL you can run or inspect
Supabase holds the daily_expenses ledger; ORM-free queries keep it fast
Prophet drives financial forecasting; custom helpers add monthly seasonality and back-testing

Challenges we ran into

Wrestling with Whisper SSL timeouts on certain VPS images (we wrote a three-tier fallback tester)
Making Prophet accept sparse SMB data; we engineered dummy-data generators and hybrid revenue/expense logic
Keeping SQL generation safe while still flexible, tight schema + constraint prompts were key
Debugging (A lot of it)

Accomplishments that we're proud of

First end-to-end run: voice memo → SQL insert → forecast dashboard in <30 seconds
Forecast MAPE under 15 % on back-tests
Zero-click onboarding: paste Supabase keys and go.

What we learned

Forecasts are trusted only when you show the error bars — transparency drives adoption.
SMB owners value actionable sentences more than charts; summarization matters as much as modeling.
Bulletproofing audio pipelines requires old-school resilience patterns (timeouts, fallbacks, logs).