Teams about to fine-tune
You have the data, but it is messy. We make it ready to train on.
[ Personnn DataForge · datasets for your AI ]
Datasets to train your AI. Yours or ours.
Turn your documents, chats and tickets into a clean, checked dataset, or start from our curated ones. Either way, you know what is in it and why every example is there.
Your dataset 3 kept · 3 set aside
- Customer asks to change a delivery address
- Refund question, answered with the policy
- Refund question, answered with the policy (copy)
- Answer cites a clause that does not exist
- Reply cut off mid-sentence
- Shipping delay, apology in the company tone
Set aside, with the reason
- Refund question, answered with the policy (copy) duplicate
- Answer cites a clause that does not exist contradicts its source
- Reply cut off mid-sentence incomplete
Illustrative example
[ opening with a handful of pilot companies · waitlist ]
[ how it works ]
Two ways in, one standard.
- 01
Bring your information
Documents, chats, tickets, spreadsheets: whatever holds what your company knows. You tell us what the dataset is for.
- 02
Every example is checked
Each example is checked against its source. What does not pass is set aside with the reason: nothing is silently kept, nothing is silently dropped.
- 03
Or start from ours
Curated datasets in areas where we already did the work. See what is in one before you use it, then use it as it is or as the base for yours.
[ what you get ]
Not a pile of text. A dataset you can defend.
Your information
Made from what your company already has
Your documents, conversations and tickets become examples ready to train on, shaped for what you want the model to do.
Checked
Every example has to earn its place
Examples are checked against their source. What fails is set aside with the reason, so you can see what was left out and why.
Curated
Ready-made datasets, not scraps
Datasets we already curated for specific areas. You see what is in them before you decide to use them.
Yours to use
Open format, any training tool
You get the dataset in an open format. Train with Personnn ModelForge or with anything else you like.
[ who it is for ]
For teams that need examples they can trust.
Teams without the data
You need examples in a domain you have no material for. Start from a curated dataset.
Teams that must show their work
Legal, health, finance: you need to say where every example came from and why it is in.
Teams evaluating a model
Test a model on examples you can trust, not on whatever was lying around.
[ frequently asked ]
What people ask before joining.
What is Personnn DataForge?
A way to get datasets to train or evaluate AI models: either built from your own information, or taken from our curated datasets. Every example is checked, and what does not pass is set aside with the reason.
Can I build a dataset from my own information?
Yes. You tell us what the dataset is for and where your information lives, and we turn it into examples and check them. Your information is yours, and how it is handled is agreed with you before we start.
What do the curated datasets cover?
Areas where we already did the curation work, such as legal, privacy and compliance, cybersecurity, finance, education and Latin American knowledge. Ask for the list for your area.
How do you check quality?
Each example is checked against its source. Examples that contradict it, are incomplete or duplicate another are set aside with the reason, so nothing is silently kept or silently dropped.
What format do I get?
An open format that works with Personnn ModelForge and with any other training tool.
Is it available now?
Not yet as a self-serve product. We are opening the first pilots through a waitlist: tell us what your dataset is for.
[ start ]
Tell us what your dataset is for.
Leave your email and one paragraph: what the dataset should teach or test, and where your information lives today (docs, chats, tickets, whatever). We write back to scope it together.
No commitment: if your data is not ready, we tell you what is missing and how to prepare it. We are opening the first pilots now.