Amin FollowupOS AI

Turn scattered WhatsApp sales chats into clear, timely follow-up actions.

Comment

Inspiration

Microbusiness owners already receive valuable sales enquiries through WhatsApp, but many opportunities are lost because conversations become buried, follow-ups are inconsistent, and sellers cannot quickly decide which prospect deserves attention.

Amin FollowupOS AI was created to help sellers recover value from leads they already have instead of constantly searching for more leads.

What it does

Amin FollowupOS AI converts a pasted WhatsApp sales conversation into:

  • a structured lead snapshot;
  • Hot, Warm, or Cold classification;
  • the reason for the classification;
  • an opportunity value when supported by the conversation;
  • recommended follow-up timing;
  • one clear next action; and
  • a concise, ready-to-send follow-up message.

The system analyses the prospect’s latest intent, need, budget signal, urgency, and main objection.

How we built it

The application uses a responsive HTML, CSS, and JavaScript frontend hosted on Netlify.

A Netlify serverless function sends the pasted conversation to the OpenAI Responses API using GPT-5.6 Terra and strict structured output.

The AI response is returned as structured JSON so the interface can reliably display the lead snapshot, temperature, opportunity value, follow-up timing, next action, and suggested message.

Deterministic business rules reinforce critical sales situations. For example, when the required item is unavailable, the customer rejects alternatives, and only requests a restock notification, the system classifies the lead as Cold for now.

Codex was used to audit the repository, review the classification workflow, troubleshoot deployment issues, verify secret handling, improve the judging path, and prepare the submission documentation.

The OpenAI API key remains inside the serverless environment and is never exposed in frontend code. All demonstration conversations are synthetic.

Challenges we ran into

The main challenge was ensuring consistent lead classification rather than depending entirely on probabilistic model judgement.

During testing, an unavailable product could still be classified as Warm because the prospect remained interested. However, no sale could progress until the required stock returned.

We solved this by combining GPT-5.6 analysis with deterministic business rules for critical sales states.

We also refined the prompt to prevent invented prices, reduce overly long WhatsApp messages, detect implicit objections, and recommend appropriate follow-up timing.

Accomplishments that we're proud of

  • Built and deployed a working end-to-end AI MVP.
  • Validated Hot, Warm, and Cold scenarios twice.
  • Extracted grounded opportunity values without inventing prices.
  • Generated concise, context-aware WhatsApp follow-up messages.
  • Protected the API key using server-side environment variables.
  • Combined structured AI output with deterministic business rules.
  • Produced a responsive interface for desktop and mobile use.

What we learned

AI becomes more useful for operational workflows when model reasoning is combined with:

  • strict structured output;
  • explicit business rules;
  • realistic synthetic test data;
  • clear acceptance criteria; and
  • disciplined MVP scope.

We also learned that a useful sales assistant should not chase every prospect. It must recognise when immediate action is appropriate and when the seller should wait for a meaningful future trigger.

What's next for Amin FollowupOS AI

The next phase will add:

  • a visible Action Queue dashboard;
  • persistent user accounts;
  • scheduled reminders;
  • controlled WhatsApp integration;
  • lead-history tracking;
  • multi-user workspaces; and
  • analytics for recovered sales opportunities.

The long-term vision is to help microbusiness owners turn scattered WhatsApp conversations into a structured and measurable sales recovery workflow.

Built With

Share this project:

Updates