The 5-Job Sorting Kit

Set up Laya (github.com/NandhaKishorM/laya) in this folder for the user. Install it with pip install laya, load it with the Router so it picks the right checkpoint per message, and create a file called sort.py with one function that takes a message and a question set and returns the answers with their confidence. Test it on one sample email and show the user the output. Explain each step in plain language.

Using sort.py, create a question set called inbox_triage for the user. Choice question "What should happen with this email?" with options: reply_now (a customer or partner waiting on us), reply_later (useful but not urgent), ignore (newsletters, promos, cold pitches). Add a score question "How urgent is this?" with options not urgent, soon, blocking. Run it on the 20 emails in the inbox_samples folder and print a table of subject, decision and confidence.

Using sort.py, create a question set called lead_score for the user. Choice question "How ready is this lead to buy?" with options: hot (asks about price, timing or a call), warm (interested but no timeline), cold (just browsing or not a fit). Add a yes/no question "Does this lead mention a budget?" Run it on the leads.csv file and add two new columns with the answers, sorted hot first.

Using sort.py, create a question set called review_alert for the user. Yes/no question "Is the customer angry or threatening to leave?" and a score question "How serious is the complaint?" with options minor, real problem, urgent. Run it on reviews.csv and print only the reviews where the answer is yes, most serious first.

Using sort.py, create a question set called support_route for the user, based on the Laya README example. Choice question "Which department should handle this?" with options: billing (invoices, payments, refunds), technical (bugs, outages, system errors), sales (pricing, new contracts), other (everything else). Add a yes/no question "Does the user threaten to cancel or leave?" Run it on tickets.csv and save a routed_tickets.csv file.

Using sort.py, create a question set called dm_filter for the user. Choice question "What kind of message is this?" with options: buyer (asks about a product, price or how to work with us), fan (compliment or general comment), collab (brand or creator partnership), spam (bots, scams, links). Run it on dms.csv and print the buyers first so the user can answer them today.

The user has a CSV of past messages with the correct label for each one (column "text" and column "label"). Prepare it for the Laya fine tuning notebook on Kaggle: clean it, balance the labels, split it into train and test, and write a short checklist for the user explaining how to upload it to Kaggle and run the notebook on the free T4 GPUs. Aim for at least a few thousand labeled examples.