Mahal Systems SalesLLM
Fits outbound or inbound sales chat agents with no backend actions. Borrow CustomerLM, a fine-tuned customer simulator that resists role inversion, and the two-part score: an LLM judge for sales-process progress plus a BERT classifier for end-of-dialogue buying intent.
What it asks the agent to do
- move a simulated customer through a sales conversation toward a purchase decision
- handle objections and persuade without breaking role
- adapt to customer personas and difficulty levels in Chinese or English
What to borrow from it
user simulatorgrading designbaseline
Caveats
No tools or state, so it measures persuasion only; 1,000 of the 1,805 scenarios are Chinese and the authors report poor cross-lingual consistency; the simulator was trained on crowdworker conversations and 10,000 dialogues are released, so check for personal data before reuse.
Details
| Workloads | sales, finance, retail |
|---|---|
| Interaction | multi turn simulated user |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | llm judge, programmatic check |
| Size | 1,805 |
| Data | mixed |
| Personal data | possible |
| License | MIT |
| Harness | own repo |
| Status | active |