SlowSells：由 Gemma 驅動、處理繁瑣工作的銷售助理

The idea: Spend less time figuring out what needs attention in your CRM and more time moving sales forward.

Salespeople often start their day by checking leads, reviewing open deals, scanning overdue activities, and deciding who needs a follow-up. The information exists in the CRM, but turning it into a useful daily plan still takes manual effort.

I built SlowSells, an AI-assisted sales CRM that brings customer and sales management together with a daily sales brief that helps users identify their next useful action.

What I Built

SlowSells combines everyday CRM workflows with an AI assistant designed to reduce repetitive sales-planning work.

1. AI Morning Sales Brief

The AI Morning Sales Brief uses a Gemma model to generate a practical summary from relevant CRM information. Instead of starting with a blank page, a salesperson gets a clearer view of what deserves attention.

The brief is designed to turn workspace context into useful next steps rather than another generic chatbot conversation.

2. An Action Center

The dashboard highlights overdue follow-ups, deals approaching their expected closing dates, and a suggested area of focus. These signals help users prioritize their day before navigating through individual CRM sections.

3. A connected sales workspace

SlowSells includes:

Contacts and companies
Leads and deal pipeline tracking
Activities and follow-up management
Product catalogue and order management
Dashboard metrics and recent sales activity
Sales insights and AI-assisted planning

The goal is to keep the daily workflow connected instead of making users jump between disconnected tools.

4. A practical fallback

AI systems can fail or become temporarily unavailable. SlowSells includes a rules-based fallback for the Morning Sales Brief so the workflow can still provide useful recommendations when model generation is unavailable.

The AI is intended to assist with prioritization, not to replace the salesperson's judgment.

How I Built It

The application uses:

Next.js App Router for the application and server-side routes
React and TypeScript for the interface and typed application logic
Prisma for database access
MongoDB Atlas for persistent CRM data
Google OAuth for sign-in
Gemma through Google's GenAI SDK for AI-generated sales briefs
Vercel for deployment

For the Morning Sales Brief, the server retrieves relevant workspace CRM information and passes the context needed for generating recommendations to the model. The application is designed to avoid sending contact email addresses, phone numbers, and private notes in that prompt.

The generated output is presented as practical guidance, including prioritized actions and an editable follow-up draft. The fallback keeps the experience useful even when AI generation does not succeed.

Why Open Innovation Matters

I chose Gemma because I wanted to build around an open-weight model rather than make this workflow depend entirely on a closed-model API.

Open-weight models give developers access to model weights and more flexibility to experiment with model behavior and deployment approaches, subject to the model's applicable terms. That creates room to adapt an AI workflow as requirements change.

There is an important distinction in my current implementation: SlowSells uses hosted Gemma inference; it does not run the model locally or work entirely offline today. Hosting inference makes the deployed demo practical, while the open-weight model choice leaves more options for future experimentation, including compatible self-hosted deployments.

The project also keeps the AI workflow focused on relevant CRM context instead of treating every piece of customer information as necessary input.

For me, open innovation is valuable when it gives builders more choices about how intelligent features are developed, deployed, and improved.

What I Learned

Building this project reinforced that adding an AI model is only one part of building a useful AI application.

The application still needs reliable data access, workspace-aware queries, predictable output, a usable interface, and a fallback when the model is unavailable. The challenge was not just to generate text, but to connect an AI brief to a real sales workflow.

I also deployed the application to Vercel and configured Google OAuth for the production domain, making the demo accessible beyond my local development environment.

I have not measured a formal time-saving benchmark yet, so I won't claim a specific number of minutes saved. The intended benefit is to reduce the manual work required to review sales activity and decide on the next step.

Try SlowSells

Live application: https://ai-sales-crm.vercel.app

Source code: https://github.com/pratikdevelop/ai-sales-crm

I'd welcome feedback on how the daily brief could make sales workflows more useful without adding more noise to the workday.

Prize Categories

Best Use of Gemma
Best Use of MongoDB Atlas