Mahal Systems DeepPlanning
Fits booking or purchasing assistants with global budgets. Borrow the rule-based checkers (commonsense checkpoints, cart match) instead of LLM judges.
What it asks the agent to do
- plan a multi-day trip within time and money budgets using search tools
- assemble a shopping cart that satisfies cross-item constraints and coupons
- gather missing information proactively before committing a plan
What to borrow from it
grading designcase template
Caveats
Two domains only; travel itineraries are first parsed by a Qwen model before checking; license not stated.
Details
| Workloads | planning, travel, shopping, tool calling |
|---|---|
| Interaction | single turn |
| Environment | api sandbox |
| End state checked | Yes |
| Tools | 9 travel APIs, 15 shopping APIs in an offline sandbox |
| Grading | programmatic check |
| Size | 240 |
| Data | synthetic |
| Personal data | none |
| License | unknown |
| Harness | own repo |
| Status | active |