Contently AIMulti-Model Social Content SaaS

The client ran a small agency and struggled to keep up with the volume of posts required across Instagram, Twitter, and LinkedIn. Writers spent hours crafting each caption, then copied text into separate scheduling tools, which caused inconsistencies and missed deadlines. They also needed a way to enforce brand voice without hiring a copy editor for every campaign. At the same time the agency wanted a single dashboard where they could see upcoming posts, edit AI suggestions, and measure expected engagement before publishing. Their existing stack was a collection of spreadsheets and third‑party schedulers, which made reporting a manual process and introduced errors. The goal was to replace that workflow with a web app that could generate copy on demand, let users tweak it, and push the final version to the social platforms with one click.

Scope: Full-stack application, AI integration, authentication

Key features

• Generate social captions with three different LLMs and pick the highest relevance score
• Suggest image suggestions that match brand color palette
• Edit AI output directly in an inline markdown editor
• Store drafts with version history and restore any previous revision
• Schedule posts to Instagram, Twitter, and LinkedIn from a single screen
• Provide a preview of expected engagement using past performance data
• Allow custom prompt templates per campaign or client
• Authenticate users with Firebase and protect API routes with JWT
• Expose 12 REST endpoints for external integration and webhook support
• Retry failed AI calls automatically with exponential backoff
• Track content generation cost per request and display savings
• Show real‑time status of scheduled posts and any publishing errors

I built the front end with Next.js and TypeScript, using Tailwind CSS and shadcn/ui for a clean, responsive layout. Server‑side functions run on Vercel and call OpenAI, Anthropic, and a fine‑tuned Groq model, so the app can compare outputs and pick the best fit for a given brand tone. JWT tokens stored in Firebase Auth keep sessions fast and secure, while MongoDB holds drafts, version history, and scheduling metadata. To avoid hitting rate limits, I added a lightweight queue in Firebase Cloud Functions that retries failed calls with exponential backoff. The scheduling module talks to each platform's API through a unified wrapper, reducing the number of integrations from three separate codebases to a single set of 12 REST endpoints. Analytics are calculated on the fly with aggregation pipelines in MongoDB, giving users a preview of likes, comments, and reach based on historical data.

Built to a standard

• SaaS
• Multi-model
• Tested
• Responsive
• SEO optimised
• CI/CD
• High performance
• Docker
• Secure
• Accessible
• TypeScript

1. Problem

   Concurrent AI requests hit provider rate limits during peak campaign days

   Solution

   Implemented a Firebase Cloud Functions queue with exponential backoff, spreading calls over a short window and preventing throttling errors

2. Problem

   Maintaining brand voice across multiple models produced inconsistent tone

   Solution

   Added a relevance scoring step that compares each model's output against a brand guideline dataset stored in MongoDB, then selects the highest‑scoring copy

3. Problem

   User authentication had to work across web and mobile without re‑login

   Solution

   Used Firebase Auth with JWT refresh tokens, allowing seamless session continuation on any device

4. Problem

   Scheduling posts required handling three separate platform APIs with different rate limits and data formats

   Solution

   Created a unified wrapper that normalizes requests and responses, then maps them to each platform's endpoint, reducing code duplication