*This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
*
My sister is an interior designer. She's brilliant at her work ,transforming living rooms, selecting palettes, managing clients. But every time she finishes a project, she spends another hour staring at her phone trying to write captions for Instagram, LinkedIn, and Twitter.
Three different platforms. Three different tones. Same project. Every single time.
So this weekend I built her something.
What I Built
LogToPost — an AI agent that turns a 2-line activity log into platform-ready social media content.
She types: "Finished designing a living room for a client in Indore. Used earthy tones throughout."
The agent does the rest.
Instagram: Warm, casual, emoji-friendly with hashtags
LinkedIn: Professional storytelling with a question that invites comments
Twitter: Punchy, under 280 characters
But it doesn't just generate once and hand it over. It runs a 4-step reasoning pipeline:
Step 1: Analyze — Extract key moments, tone, target audience, emotional hook
Step 2: Generate — 3 options per platform using different content angles
Step 3: Critique — AI reviews its own output: "Does this sound human or like AI wrote it?"
Step 4: Refine — Improves the best option with one specific change
That last step is what makes it feel like an agent, not a chatbot. It generates, reflects, and improves, without being asked twice.
Why Open Source AI
I used Groq's inference API running GPT-OSS 120B (OpenAI's open-weight model) — not ChatGPT, not Claude.
Here's why that matters for my sister specifically:
Her client work is private. Every log she types contains real client names, real project details, real business information. I wasn't going to send that to a closed API where it might be used for training.
I can change how the agent behaves. If she says "the Instagram captions feel too generic," I edit the prompt and redeploy. No waiting for a model update. No paying for fine-tuning. Just open, editable logic.
It costs nothing to run. Groq's free tier handles all the inference. She gets a personal AI content team for zero ongoing cost.
With open-weight models, her data stays in the pipeline I control. If Groq changes their pricing tomorrow, I swap to Ollama locally in an afternoon. That flexibility is only possible because the models are open.
How It Works
The agent runs 6 LLM calls in sequence - one to analyze, three to generate platform-specific options, three to critique and refine its own output. Each call builds on the previous one. That's what makes it feel like reasoning, not just autocomplete.
The critique prompt specifically asks: "Does this sound human or like AI wrote it?", which is the one question that matters most for social media content.
The Tech Stack
- Backend: Node.js + Express
- Database: MongoDB Atlas
- AI: Groq API running GPT-OSS 120B (open-weight)
- Frontend: React with TanStack Router
- Deployment: Render (backend) + Lovable (frontend)
What She Said
I handed it to my sister. She typed one line about a reading nook she'd just finished for a client.
The Instagram post it generated started with: "We just finished this cozy reading corner! Swapped the bare wall for a warm walnut accent and added a custom leather-soaked reading chair..."
She read it, looked up, and said: "This sounds like me."
That's the only metric that matters.
Why Open Innovation Matters Here
If I'd built this on ChatGPT's API:
- Her client data goes to OpenAI
- I can't change how the model reasons without paying for fine-tuning
- I'm locked into one provider's terms forever
With open-weight models:
- Data stays in my pipeline
- I edit prompts and redeploy in minutes
- Free tier covers everything she needs
- If one provider changes, I swap models in an afternoon
Open source isn't just a technical choice here. It's the reason I could build something trustworthy for someone I actually care about.
What's Next
- Voice input: log your day with a voice memo (Whisper API)
- Video support: extract key moments from project walkthrough videos
- Direct posting: approve and publish straight to Instagram and LinkedIn
Links
Built in one weekend for my sister, who deserves to spend her time designing , not writing captions.
Top comments (0)