SignalDesk: A Privacy-First Customer Feedback Desk for Small Businesses

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

SignalDesk is a privacy-first customer-feedback intelligence desk for small service businesses: gyms, clinics, studios, agencies, and subscription businesses.

These businesses already have survey or CRM exports, but usually do not have a data team. SignalDesk turns a CSV into:

• Risk-ranked customers
• At-risk customer segments
• Satisfaction and wait-time signals
• A short retention brief for the business owner

The product is intentionally small and specific: a business owner can run a weekly feedback export through it instead of manually scanning rows in a spreadsheet.

Demo

The current version runs locally:

python app.py

Then open http://127.0.0.1:8765 and upload the included sample CSV.

A public deployment is the next step. The project is still useful without a cloud deployment because privacy-sensitive businesses can keep the workflow on their own laptop.

Code

SignalDesk on GitHub

How I Built It

SignalDesk is a small Python web application using only the standard library for its first runnable version. It accepts CSV data, validates the required feedback fields, calculates a transparent local risk baseline, groups results by segment, and generates a retention brief.

The app has an optional local Gemma integration through Ollama. When Ollama and a Gemma model are available, the brief is generated locally. When they are not available, SignalDesk uses a deterministic fallback so the core workflow remains usable.

There is also an explicit TabPFN adapter seam for upgrading the transparent baseline to a tabular foundation model when the optional Prior Labs dependency and credentials are configured.

Why Does Open Innovation Matter?

Customer feedback can contain names, complaints, health-related context, or commercially sensitive information. Sending every export to a hosted AI API is not always acceptable for a small business.

Open models make a different product possible:

• Gemma can run on the business's own laptop through Ollama.
• The raw CSV does not need to leave the machine.
• The business is not locked into per-token pricing.
• The model can be swapped or fine-tuned for a specific industry.
• The deterministic fallback makes the workflow inspectable instead of hiding every decision behind an opaque API.

The trade-off is real: local inference requires suitable hardware and may be slower. SignalDesk therefore separates analysis from narrative generation and clearly labels which model produced the brief.

My Agent Session

This project was built iteratively with an AI coding assistant, including the local web app, CSV analysis workflow, optional Gemma integration, and free-host deployment configuration.

First User

I have not claimed a paying customer yet. The first validation target is a small service business that currently reviews customer feedback manually. The proposed pilot is a weekly report using synthetic data first, followed by a privacy-preserving local run with the owner's own export.

Prize Categories

• Best Use of Gemma — optional local Gemma generation through Ollama
• Best Use of TabPFN — adapter prepared for tabular risk prediction; full integration is the next implementation step

I will only claim a partner category after verifying the corresponding technology in the submitted build.