Spaces:
Sleeping
Sleeping
File size: 1,776 Bytes
e6a02dd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 | spec_version: 1
name: salespath
type: space
runtime: fastapi
app: salespath_env.server.app:app
port: 7860
description: >
SalesPath is an OpenEnv-compatible RL environment for training LLM
agents to navigate a multi-step B2B sales workflow. The agent must
PROSPECT, QUALIFY, PRESENT, HANDLE_OBJECTION, OFFER_DEMO, NEGOTIATE,
CLOSE or DISQUALIFY while obeying nine business rules verified
programmatically by a deterministic rule-based ProspectSimulator
(no LLM in the verifier).
action_space:
type: structured
schema:
action_type:
type: enum
values:
- PROSPECT
- QUALIFY
- PRESENT
- HANDLE_OBJECTION
- OFFER_DEMO
- NEGOTIATE
- CLOSE
- FOLLOW_UP
- DISQUALIFY
content:
type: string
target:
type: string
observation_space:
type: structured
fields:
prospect_response: string
workflow_stage: string
constraints_violated: list[string]
steps_completed: list[string]
turn_number: int
reward: float
reward_components: dict
done: bool
rubric:
type: weighted_sum
components:
- name: outcome
weight: 0.20
- name: compliance
weight: 0.40
- name: ordering
weight: 0.20
- name: efficiency
weight: 0.10
- name: format
weight: 0.10
difficulty_levels:
- level: 1
description: Budget known, decision-maker present, easy close
- level: 2
description: Budget hidden, one objection, demo required
- level: 3
description: Budget hidden, two objections, possible stalling
- level: 4
description: Adversarial — misleading signals, correct action is DISQUALIFY
theme: long_horizon_planning_and_instruction_following
bonus_track: scale_ai_business_workflows
|