AI-Driven Planning System for Case Workers

Category	Details
Industry	Non-Profit, Social Services
Company Size	Mid-Sized (100–499 employees)
Location	United States (multi-state operations)
Goal	Improve social mobility outcomes by enabling faster, more consistent, and high-quality individualized planning while reducing administrative burden and ensuring secure, state-compliant operations.
Technologies Used	• LangChain • LangGraph • LangFuse • Mermaid-based decision tree • Deepgram • Google Cloud Speech-to-Text for audio transcription and diarization

Overview

A US-based non-profit working with public-sector agencies engaged Greyhaven to develop a new planning product designed to improve social mobility outcomes. The goal was to strengthen how individuals are supported as they transition into new community settings by equipping social workers and case managers with a platform that enables faster, more consistent, and better-informed planning.

Greyhaven partnered with the organization to design and build a secure, AI-driven planning system that transforms fragmented intake information into clear, personalized support plans. The system is designed to handle complex, evolving individual needs while adapting to state-level programs, policies, and administrative requirements. Plans can be updated over time and deployed at scale across multiple jurisdictions.

The Challenge

Social workers and case managers often manage hundreds of concurrent cases, each requiring:

• Rapid assessment of housing, employment, healthcare, and support needs
• Identification of appropriate local and regional resources
• Construction of multi-month support plans with defined milestones
• Ongoing reassessment, as individual circumstances change

Much of this work must begin early in the engagement lifecycle, when time is limited and information is incomplete.

Historically, this planning has relied on manual processes, individual experience, and disconnected tools. As a result, planning was slow, outcomes varied widely, and it was difficult to apply consistent approaches across jurisdictions with different policies and service models. Greyhaven was engaged to replace this fragmented workflow with a secure, AI-driven planning system that supports faster, more consistent decision-making at scale.

Why a Custom AI System Was Required

Off-the-shelf case management software and generic AI tools were insufficient.

Key requirements included:

• State-by-state variability in social service programs and administrative rules
• Intake workflows that differ based on timing, location, and agency context
• Highly sensitive personal data, requiring strict control over hosting and access
• Decision support that depends on reasoning and synthesis, not simple automation

This problem required a bespoke AI system, purpose-built for public-sector constraints and real-world operational complexity.

The Greyhaven Solution

Greyhaven designed and deployed a production-grade, AI-driven planning platform that transforms intake information into structured, actionable support plans while operating within complex, state-specific constraints.

Core Architecture

• Built on LangChain and LangGraph to model each case as a multi-step reasoning workflow
• Encodes jurisdiction-specific rules and program requirements directly into the orchestration layer
• Uses Mermaid-based decision trees to formalize branching logic and maintain policy-aligned reasoning
• Produces structured outputs (action items, milestones, timelines, responsibilities) rather than free-form text

Production Observability

• Instrumented with LangFuse for prompt tracing, execution monitoring, and version control
• Enables regression detection, controlled iteration, and safe deployment in regulated environments

Intake & Transcription Pipeline

• Integrated Deepgram and Google Cloud Speech-to-Text for secure transcription and speaker diarization
• Converts live or recorded interviews into structured case variables
• Feeds transcript-derived data directly into the LangGraph planning workflow

Deployment Model

• Built API-first (to separate reasoning services from the user interface)
• Supports modular scaling and external system integration
• Initially deployed on Greyhaven-managed infrastructure, then migrated to the client’s controlled environment, at their preference

The system combines advanced reasoning with modern intake tooling to greatly reduce manual work and accelerate planning.

Results and Impact

Following deployment, the organization achieved:

• Substantial time savings (30%+) for caseworkers creating individualized plans
• The ability to begin planning earlier in the engagement process
• More consistent, higher-quality plans across staff and jurisdictions
• A shift from static plans to living, continuously updated evaluations
• A scalable foundation for expansion into additional states

What began as a proof of concept evolved into a core product capability that supports organizational growth and mission delivery.

Ongoing Partnership

The platform continues to evolve as the organization expands into new jurisdictions and adapts to new policy and program requirements. Greyhaven supports:

• Architectural improvements
• Feature expansion
• Scaling across states
• Advanced AI and reasoning enhancements

The organization operates the platform independently, day to day, while relying on Greyhaven for continued evolution and long-term strategy.

Why This Matters

This engagement exemplifies Greyhaven’s approach to applied AI:

• Bespoke systems where generic tools fall short
• Deployment in high-sensitivity, regulated environments
• Architectures that prioritize data sovereignty and institutional trust
• AI designed to improve real-world decision-making, not just automate workflows

This case study demonstrates how carefully engineered AI can deliver measurable impact in domains where correctness, security, and accountability matter most.