- ▹Role: Solo Founder
- ▹Product: Raccog website and app platform
- ▹Focus: Python APIs, document-based RAG, generation, previews, and publishing
- ▹Stack: Python, FastAPI, Next.js, React, TypeScript, Node.js, MongoDB, AWS S3
- ▹Website: raccog.com
I built Raccog as a solo founder. It is a website and app platform with AI-powered generation, previews, publishing, analytics, and workflow features. My work includes Python backend APIs with FastAPI and document-based retrieval-augmented generation, or RAG.
The AI work is part of a wider product. I also built with Next.js, React, and TypeScript, used MongoDB and S3-backed storage, and implemented the systems needed to preserve generation work and publish it.
The product behind the AI feature
Raccog's public product centers on describing what to build, refining it, and publishing the result. Supporting that experience requires application state and delivery features around the generated content.
For this project, my engineering scope includes persistent generation jobs, sandboxed React previews, database-backed publishing, and release snapshots. Those components handle different stages of the work, from generation in progress to a published version.
As the solo founder, I work across those stages and the APIs, interface, and storage that connect them.
Python backend APIs with FastAPI
I developed Python backend APIs using FastAPI. This is an explicit part of my Raccog work alongside the platform's JavaScript and TypeScript components.
FastAPI provides an API layer for the Python functionality. That work is part of the backend, connected to a wider product with its own frontend, application state, and publishing features.
Raccog combines Python/FastAPI with Next.js, React, TypeScript, and Node.js components. Building across those technologies means working on the boundaries between services as well as the implementation within them. An AI capability needs an interface the rest of the application can use, and the application needs to represent the work happening behind that interface.
Document-based retrieval-augmented generation
I implemented document-based RAG workflows. RAG uses retrieved information as context for a model's response. In this work, documents provide that information source.
This gives the workflow a way to use information from documents during generation. Retrieval is the step that brings relevant source material into the model's context; the model then works with that context as part of the application workflow.
My contribution covers the document-based RAG implementation and the Python backend APIs. It is one part of Raccog's AI engineering, alongside the wider product work needed to generate, preview, and publish websites and applications.
Generation jobs and application state
I implemented persistent generation jobs. Persistence gives the product a stored record of generation work rather than leaving the work only in the current interface session.
That record is part of the broader application state. The platform uses MongoDB and S3-backed storage, alongside its frontend and backend components. These pieces support the product around generation, rather than treating the generated output as the entire application.
Generation jobs are therefore part of the product's data model and backend work. The interface presents an experience to the user, while the application also needs records of the work happening behind it. Persistent jobs connect those concerns.
Previewing the result
I implemented sandboxed React previews so the generated application can be inspected in a preview environment. That makes previewing a distinct stage of the product workflow.
A preview is useful because generated output needs to be seen in the form in which people will interact with it. The preview system is therefore part of the application-building experience, alongside the generation itself.
Working on the preview also means connecting the generated output to an application runtime and then back to the interface. This is where my frontend and backend responsibilities meet: the result has to be represented as a running React application within the builder's experience.
Publishing and release snapshots
I implemented database-backed publishing and release snapshots. Publishing moves the product into a different stage: there is a version intended to be available outside the builder.
Release snapshots record a release state. The distinction between ongoing work and a recorded release is useful when explaining how an application-building platform is structured.
For Raccog, I owned both the database-backed publishing implementation and the release snapshots. These features extend the platform beyond generating an initial application into managing a version intended for publication.
The wider platform
I also developed analytics and workflow features. These sit alongside generation, previews, and publishing as part of the broader Raccog product.
The product spans several connected systems: Python APIs and document-based RAG, the web application, storage, generation jobs, previews, and publishing. Analytics and workflow features extend that product beyond the initial building experience.
What the project demonstrates
Raccog shows my ability to combine AI functionality with full-stack product delivery. I own the work across the Python API layer, the web application, and the systems around generation and publishing.
The result is a live website and app platform that can be visited publicly. The project brings together my Python and FastAPI work, practical RAG implementation, and full-stack product ownership in one application.
Explore the product
- ▹Visit Raccog: raccog.com
- ▹Discuss a project: rahul.raccog.com/contact