Today, we're launching Promptrepo on Product Hunt. Sign up through Product Hunt and get 3 months free.

Last week, Kevin Weil, OpenAI's Chief Product Officer, shared how fine-tuning has become central to OpenAI's product development:

  • They deploy ensembles of fine-tuned models for customer support and workflow automation. So, when you ask a question in their support forum, a fine-tuned model is answering you.
  • For ChatGPT's deep research features, they created custom evaluation benchmarks and iteratively fine-tune models until performance targets were met.

In short, they use fine-tuning for everything—from support to core product development. He predicted that AI teams will soon embed dedicated researchers to fine-tune models using organization-specific data. Prompting alone, he emphasized, isn't enough for high-stakes use cases.

I was thrilled to hear him praise fine-tuning. For years, I've been urging my team to adopt fine-tuning over prompting. This felt like long-awaited validation. And it's not just my team—most developers I talk to still stick with prompting. So, what's holding them back?


Why Developers Avoid Fine-Tuning?

Despite its benefits, fine-tuning hasn't gone mainstream. Here's why:

  • Technical Complexity — Requires specialized formats (like JSONL), consistent labeling, and custom scripts.
  • Slow Feedback Loops — Unlike prompting, you can't see immediate results.
  • Deployment Friction — Training and serving fine-tuned models demand infrastructure and monitoring.
  • Expertise Bottlenecks — Product managers and domain experts often have the examples for fine-tuning—but they need engineers to execute it.

These challenges lead teams to stick with prompting—even when results are inconsistent and the system becomes fragile.


What If Fine-Tuning Was as Easy as Prompting?

We faced these same hurdles during COVID. Small businesses were using Google Forms to take orders, and we built an add-on to calculate totals. But prices were buried in field descriptions, with wildly inconsistent formats. So, we used prompting to extract prices.

It worked—until it didn't. Currency symbols, discounts, and bulk pricing caused failures. When OpenAI launched its fine-tuning API, I compiled pricing examples into a spreadsheet and trained a model. Accuracy improved dramatically. But I still had to update data, retrain, and redeploy every time something changed.

That didn't scale. Our support lead saw the issues firsthand, but couldn't fix them. So, we built a Google Sheets add-on to make fine-tuning accessible for him. Now, our support lead could:

  1. Add training examples in a spreadsheet
  2. Trigger training with one click
  3. Deploy the updated model without writing code
  4. Validate results immediately


How We Use Fine-Tuning in Our Products

This spreadsheet-driven fine-tuning system now powers all the AI workflows across our products:

  • Formesign (Form Generation): Turns form titles and scanned documents into structured forms using fine-tuned template models.
  • Neartail (Payment Processing): Parses UPI payment SMS messages, matches them to orders, and auto-updates payment status—saving merchants from card fees.
  • Near.tl (Food Intelligence): Crawls restaurant websites to classify food types, extract nutritional data, and compute NutriScores.

After years of internal use, we're now opening this tool to developers and product teams. If you're working with LLMs and struggling with:

  • Inconsistent model behavior
  • Prompt chains that break in production
  • A need for experts to guide AI without involving engineering
  • Difficulty scaling AI features


Try Promptrepo: Fine-Tuning for Everyone

Promptrepo brings the power of fine-tuning to your entire team through a familiar interface: Google Sheets. Under the hood, our platform handles all the complex parts:

  • No-code AI Model Building — Domain experts can create and refine AI models directly in spreadsheets without coding knowledge
  • Multi-Model Support — Choose between OpenAI, Mistral, or LLaMA for training and deployment with just a few clicks
  • Built-in Evaluation — Test models easily using Google Sheets formulas, our intuitive UI, or API calls
  • Versatile Model Types — Build classification, extraction, or generative AI models tailored to your specific use cases
  • Seamless API Integration — Connect your fine-tuned models to apps, websites, and workflows through our API
  • Complete Version Control — We handle everything from build process to versioning, letting you focus on accuracy

The result? Domain experts can dramatically improve model performance—without writing a single line of code.


Let's Make Fine-Tuning Routine

Fine-tuning shouldn't be reserved for AI labs like OpenAI. It should be part of every product team's toolbox—just like a spreadsheet. This is how AI becomes truly useful: when anyone on your team can teach it.

Today, we're launching Promptrepo on Product Hunt. Sign up through Product Hunt and get 3 months free.

Try it. Break it. Tell us what's missing. Let's make fine-tuning routine—for everyone.