Inspiration
I have worked in ecommerce consulting for more than a decade. Long before ChatGPT existed, I wanted to build a system that could bring every part of my work together.
The difficult part of ecommerce consulting is not always knowing what to do. The difficult part is collecting reports, organizing data, building dashboards, reviewing performance, documenting recommendations, and repeating that process across multiple clients and marketplaces. Valuable time that could be spent making strategic decisions is instead consumed by administrative work.
My goal was never to remove the consultant from the process or let software make generic decisions without context. I wanted to create a system that could carry my knowledge, organization, and recommendations across many clients while keeping me in control of the actual strategy.
That idea became SellerPilot, an ecommerce client terminal that combines my public business website, report import tools, private client workspaces, operational dashboards, and an administrative console into one connected platform.
What it does
SellerPilot gives each authorized client a private workspace where they can review the information that matters to their ecommerce business.
The platform can import Amazon reports and organize them into dashboards for:
- Sales
- Transactions
- Returns
- Restock recommendations
- Advertising performance
- Search term performance
Clients can review their reporting history and ongoing work without searching through disconnected emails, spreadsheets, attachments, and documents.
SellerPilot also includes an administrative workspace where I can create and manage client access, control memberships, and organize each client separately. The system is invitation only, and its data model is designed to prevent one client from accessing another client's information.
The public SellerPilot website explains the service, while the interactive demo allows visitors to explore a fictional version of the client terminal without creating an account or accessing real client data.
How I built it
I began using GPT-5.6 as soon as it became available. When ChatGPT Sites launched, I used it to migrate my public website from WordPress and turn it into a more modern representation of SellerPilot.
I then used Codex to build the larger application behind the website. Over approximately one week, the project expanded into a functioning client portal, report importer, administrative console, database structure, access control system, and collection of ecommerce dashboards.
SellerPilot uses JavaScript, HTML, CSS, Node.js, Firebase, and Google Cloud. Firebase Authentication handles access, Cloud Storage receives uploaded reports, Cloud Functions validates and processes them, and Cloud Firestore stores normalized client specific records.
The report workflow follows a structured process:
Report upload → Cloud Storage → Cloud Function → validation and normalization → client workspace → dashboard
Codex helped me turn business requirements expressed in normal language into working application architecture. GPT-5.6 helped me reason through the product, define the workflows, troubleshoot problems, review security boundaries, and decide how the system should behave from both the consultant and client perspectives.
GitHub now contains a secure source backup of the platform, including the application code, Firebase configuration, security rules, tests, and architecture documentation. Credentials, production reports, and real client data are excluded.
Challenges
My biggest challenge was that I entered this project as an ecommerce consultant, not as a traditional software developer. Before building SellerPilot, I did not know what Firebase was, how Cloud Functions worked, or how a client scoped database and authentication system should be structured.
The project also became much larger than I initially expected. It was not simply a website. It required multiple hosting surfaces, report processing, authentication, invitations, permissions, data validation, security rules, administrative tools, and safeguards around real client information.
There was also a very physical limitation: my Surface 7 crashed roughly seven times while I was building the project. I learned that it was not designed for the amount of browser, cloud, coding, and testing activity I was asking it to handle. The hardware became part of the challenge, but the project survived every interruption.
What I learned
The most important thing I learned is that software development is becoming accessible to people whose expertise comes from somewhere other than programming.
I did not begin with years of coding knowledge. I began with more than a decade of ecommerce experience, a very specific operational problem, and a clear idea of how the solution should work.
GPT-5.6 and Codex allowed me to translate that professional knowledge into a real platform. Along the way, I learned about Firebase, Google Cloud, GitHub, authentication, cloud functions, data models, security rules, testing, source control, and deployment safety.
Most importantly, I learned that the system I had imagined for years was not unrealistic. It was buildable.
SellerPilot is the beginning of turning my individual ecommerce consulting process into a scalable system without removing the experience, judgment, or human strategy that gives the service its value.
Log in or sign up for Devpost to join the conversation.