From Replit MVP to Production-Ready AI SaaS Platform

Overview

RiskAssist is an AI-powered platform that enables organizations to auto-generate, edit, and manage internal policy documents such as privacy policies, terms & conditions, and compliance frameworks. It leverages advanced LLMs, tool calling, and vector search to make policy creation smarter, faster, and editable.

But the product didn’t begin this way.

Phase 1: Initial State

The original version of RiskAssist was a lightweight MVP built entirely on Replit with almost no architecture or production readiness.

Prototype Limitations:

• No authentication or access control
• No database — used localStorage only
• One static prompt hardcoded in a form
• No backend APIs or data handling
• No modular structure — all-in-one frontend code
• AI lacked prompt engineering or tool calling
• No UI design system — visually broken and unresponsive
• No version control, deployments, or environments

While the concept had promise, the product was unscalable, insecure, and far from production-grade.

Phase 2: Assessment

We conducted a complete technical audit to identify gaps and plan a rebuild from scratch. The assessment phase involved:

• Mapping missing features: backend, auth, storage, RAG, AI flexibility
• Reviewing OpenAI integration and identifying limits of static prompting
• Benchmarking frontend usability and state handling
• Defining technical goals: modular, secure, scalable, and editable AI

We aligned with the product vision and drafted a multi-phase implementation roadmap.

Phase 3: Implementation

We rebuilt the entire product using a modern, scalable tech stack optimized for AI applications.

AI Layer

• Integrated Vercel AI SDK for managing prompt flows and OpenAI tool calling
• Added dynamic prompt injection: company name, user inputs, and context
• Built fallback flows, error handling, and token limit guards
• Enabled editing and revision of AI output via prompt chaining

Vector Database & Retrieval

• Embedded industry-standard compliance frameworks into Upstash Vector DB
• Added semantic search and Retrieval-Augmented Generation (RAG) for grounded AI responses
• Linked generated policies to context from these documents

Policy Editing and Document Handling

• Integrated BlockNote markdown editor for rich, editable AI documents
• Users can regenerate parts of the document with inline AI prompts
• Enabled PDF uploads for existing policies → transformed into editable form

Auth, API, and Storage

• Used Next.js for frontend + serverless API routes
• Set up Clerk for authentication and session-based access control
• Modeled relational data with Prisma ORM and PostgreSQL
• Built RESTful APIs for policy management, file uploads, and user actions

Billing & Monetization

• Added credit-based pricing system: AI usage (generate/edit) consumes credits
• Integrated Stripe for secure payments and credit top-ups
• Admin controls and dashboard track usage, limits, and billing history

DevOps and CI/CD

• Set up GitHub version control, PR flow, and CI/CD with GitHub Actions
• Used Vercel for hosting, preview environments, and production deployments
• Configured staging and production with environment-specific secrets

Phase 4: After State

The final version of RiskAssist is a secure, full-featured SaaS platform ready for real users and investors.

Key Features:

• Full AI policy generation with prompt memory and retrieval grounding
• Rich editing interface using markdown with inline AI prompts
• User dashboard to view, manage, and update all policy documents
• Upload existing PDFs and turn them into editable policies
• Responsive, intuitive UI built with TailwindCSS
• Secure login and multi-user support with Clerk
• Stripe billing system and real-time credit tracking
• Robust backend with Prisma, PostgreSQL, and modular APIs
• CI/CD for automated deployments, previews, and rollbacks

Phase 5: Verification

We completed multiple rounds of verification before launch:

• Manual QA + automated API tests
• Role-based access testing with multiple user types
• Vector DB performance + RAG accuracy checks
• Prompt performance and fallback testing
• CI/CD with preview builds and staging/production isolation

The platform was validated across both technical and business metrics for reliability and scale.

Outcome

In just four weeks of focused engineering, RiskAssist was transformed into a production-grade AI SaaS product:

• AI-generated policies grounded in real documents
• Editable content with markdown-based UX
• Upload existing policies and edit on the fly
• Full user auth, session management, and role control
• Usage-based credit billing with Stripe
• Scalable infrastructure deployed with CI/CD

The platform is now demo-ready for legal teams, enterprise clients, and investors.

Final Thoughts

RiskAssist began as a vision coded into a Replit prototype — no backend, no auth, and no persistence. It relied on a single prompt and localStorage. The potential was clear, but the execution was fragile.

Through a full rebuild, Axentia turned the product into a robust, editable, AI-first SaaS solution. This transformation included vector-based AI grounding, tool-calling, modular APIs, CI/CD pipelines, and a live billing model.

This case study shows how the right technical execution — with the right tools and strategy — can elevate a messy MVP into a monetizable product ready for the market.