When you create a new case in The Layer, you may notice that the Category and Subcategory fields are automatically filled out after you type a subject line.
This isn't magic, it's AI assistance! šÆ
ā What It Does
As soon as you enter a subject for a new case, The Layer uses machine learning to suggest the most relevant category and subcategory based on previous cases.
Youāll see a small message under the Subcategory dropdown:
āØ The category has been selected with AI assistance. You can change it if needed.
This is there to help you move faster and reduce manual workāwhile still giving you full control.
š” Why It Helps
Saves time: No need to scroll through long lists.
Improves consistency: Keeps categories aligned across your team.
Learns from history: Based on how similar cases were categorised before.
š§¬ How It Works (Simple Version)
Under the hood, weāve trained a smart model using thousands of real case subjects from The Layer. Hereās what happens:
š You type a subject line (e.g. āLaptop won't connect to Wi-Fiā)
š§ The model converts this into a "semantic fingerprint"āa way to capture its meaning.
š That fingerprint is compared to similar past case subjects using a technique called similarity search.
š·ļø If a strong match is found, the system suggests the subcategory used in that closest case.
So itās not just keyword-matchingāit understands meaning.
How Accurate Is It?
It becomes more effective the more you use it, because it learns semantic meaning, not just exact words.
š ļø Behind the Scenes (For the Curious)
The system uses a fine-tuned SentenceTransformer model.
It stores vector āembeddingsā of all previous case subjects.
When a new case is added, the model finds the closest match using cosine similarity.
The matched caseās subcategory is suggested to you in real-time.
All of this runs securely in your environment, with no manual intervention needed.
š§ Can I Change the Category?
Absolutely. Think of this as a helpful suggestion, not a final decision.
If the category isnāt quite right, just change itāyour input helps the system get better over time.
š Whatās Next?
Weāre working on:
Even faster suggestions using FAISS (for nerds: itās like Google for vectors)
Better handling of tricky cases (like similar software subcategories)
If you have feedback or questions about category suggestions, reach out to our product team,weāre all ears!