# CRM Data Pipeline ## Docs - [CRM Data Pipeline: OpenEnv Benchmark for AI Agents](https://ritika-fa61c43f.mintlify.app/introduction.md): The CRM Data Pipeline Environment simulates real enterprise data engineering workloads for evaluating autonomous AI agent capabilities. - [Environment Architecture and Data Flow Explained](https://ritika-fa61c43f.mintlify.app/architecture.md): A deep-dive into the CRM Pipeline's POMDP model, server components, file interconnections, reward signals, and full episode lifecycle. - [Task Catalogue: Three Difficulty Tiers of CRM Cleaning](https://ritika-fa61c43f.mintlify.app/task-catalogue.md): Detailed breakdown of all three benchmark tasks — Easy, Medium, and Hard — with data schemas, objectives, and scoring criteria. - [CRM Data Pipeline Environment: Get Started in 5 Minutes](https://ritika-fa61c43f.mintlify.app/quickstart.md): Spin up the CRM Data Pipeline environment server and run your first agent episode loop with real data cleaning actions in under 5 minutes. - [CRM Data Pipeline: Local and Docker Installation Guide](https://ritika-fa61c43f.mintlify.app/installation.md): Install the CRM Data Pipeline Environment locally with uv or pip, or run it as a containerized service using the included Dockerfile. - [CRM Data Pipeline Environment Configuration Reference](https://ritika-fa61c43f.mintlify.app/configuration.md): Set environment variables, choose task difficulty, configure your LLM provider, and tune agent behavior for the CRM Data Pipeline benchmark. - [POMDP Model: How the Environment Simulates Data Engineering](https://ritika-fa61c43f.mintlify.app/concepts/pomdp-model.md): Learn how the CRM Pipeline models data cleaning as a Partially Observable Markov Decision Process with episodic state, transitions, and terminal scoring. - [Action Space: All 8 Pipeline Operations Explained](https://ritika-fa61c43f.mintlify.app/concepts/action-space.md): Reference guide for all 8 CRMPipelineAction types — required fields, reward values, JSON examples, and enum values for every pipeline operation. - [Observation Space: What Your Agent Sees After Each Step](https://ritika-fa61c43f.mintlify.app/concepts/observation-space.md): Understand the CRMPipelineObservation object — its fields, how they update after each action, and how to use them to guide your agent's decisions. - [Reward System: Dense Heuristic Reward Shaping Explained](https://ritika-fa61c43f.mintlify.app/concepts/reward-system.md): How the CRM Pipeline's dense heuristic reward system guides agents through progressive cleaning steps and penalizes destructive or premature actions. - [CRM Data Sources: Schemas, Formats, and Corruption Patterns](https://ritika-fa61c43f.mintlify.app/data-engine/data-sources.md): Reference for all four CRM data sources — web_forms, salesforce, web_leads, and legacy_db — schemas, column types, and per-source corruption patterns. - [Data Generator: How Messy CRM Data Is Procedurally Created](https://ritika-fa61c43f.mintlify.app/data-engine/data-generator.md): Deep-dive into the data_generator.py chaos engine — how base truth is generated, corrupted, and used to create realistic messy CRM benchmark datasets. - [Standardization Strategies for CRM Column Cleaning](https://ritika-fa61c43f.mintlify.app/data-engine/standardization-strategies.md): How to use all four standardization strategies — TO_DATETIME_ISO, LOWERCASE_STRIP, EXTRACT_NUMBERS, and PHONE_E164 — to clean corrupted CRM columns. - [CRM Deduplication Strategies: Exact and Fuzzy Matching](https://ritika-fa61c43f.mintlify.app/data-engine/deduplication.md): Use EXACT_EMAIL and FUZZY_NAME_PHONE deduplication strategies to eliminate duplicate CRM records inserted by the data generator's chaos engine. - [SQL Execution: Querying CRM DataFrames with EXECUTE_SQL](https://ritika-fa61c43f.mintlify.app/data-engine/sql-execution.md): Run SQL SELECT queries over CRM DataFrames using EXECUTE_SQL — covering joins, COALESCE conflict resolution, bot filtering, and safety guardrails. - [Running the Baseline Inference Script on CRM Tasks](https://ritika-fa61c43f.mintlify.app/guides/running-inference.md): Execute the built-in inference.py script to run an LLM agent or rule-based baseline across all three CRM cleaning tasks and evaluate scores. - [Write a Custom Agent for the CRM Data Pipeline Env](https://ritika-fa61c43f.mintlify.app/guides/writing-an-agent.md): Build a custom Python agent that interacts with the CRM Pipeline REST API using the typed client library, action models, and observation parsing. - [Understanding Grading and Scoring in the CRM Pipeline](https://ritika-fa61c43f.mintlify.app/guides/grading-scoring.md): Learn how the CRM Pipeline grader evaluates your agent's output against the hidden ground truth, including the row-match formula and excess-row penalty. - [Validate Your Submission Before Submitting to OpenEnv](https://ritika-fa61c43f.mintlify.app/guides/submission-validation.md): Use the validate-submission.sh script to run the 3-step pre-submission check: HuggingFace Space ping, Docker build, and openenv validate. - [CRM Data Pipeline REST API: Complete Endpoint Overview](https://ritika-fa61c43f.mintlify.app/api/overview.md): Overview of all five REST endpoints exposed by the CRM Data Pipeline server, base URL conventions, and JSON request/response format standards. - [POST /reset — Initialize a New CRM Pipeline Episode](https://ritika-fa61c43f.mintlify.app/api/reset.md): POST /reset initializes a fresh episode, loads messy CRM data for the configured task, and returns the first observation with task objective and sources. - [POST /step — Execute a Pipeline Data Engineering Action](https://ritika-fa61c43f.mintlify.app/api/step.md): POST /step executes a CRMPipelineAction on the active episode, mutates in-memory DataFrames, and returns an observation with reward and feedback. - [GET /state — Retrieve Current Episode State Metadata](https://ritika-fa61c43f.mintlify.app/api/state.md): GET /state returns the current CRMPipelineState including the episode UUID, step count, and active task ID for tracking agent progress. - [POST /grader/{episode_id} — Score a Completed Episode](https://ritika-fa61c43f.mintlify.app/api/grader.md): POST /grader/{episode_id} compares your agent's final DataFrame against hidden ground truth and returns a normalized score from 0.01 to 0.99. - [CRMPipelineAction: Agent Action Payload Model Reference](https://ritika-fa61c43f.mintlify.app/api/models/action.md): Complete field reference for CRMPipelineAction — the Pydantic model sent in POST /step, covering all 13 fields and which action_type requires each. - [CRMPipelineObservation: Parsing Agent Observations](https://ritika-fa61c43f.mintlify.app/api/models/observation.md): Complete reference for CRMPipelineObservation — the response model from POST /step and POST /reset, with field semantics and parsing examples. - [Enum Reference: All Five CRM Pipeline Strategy Enums](https://ritika-fa61c43f.mintlify.app/api/models/enums.md): Reference for all five CRM pipeline enums — PipelineActionType, StandardizationStrategy, MissingStrategy, DeduplicationStrategy, and ConflictRule. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.