AI-Powered School Improvement Plan Review System
An EdTech Company
Client
Challenge
- Manual Review Bottleneck: The client’s operations team faced a time-consuming manual review process for extensive school improvement plans (60-100 pages each), requiring meticulous verification of federal compliance and educational best practices alignment.
- Regulatory Compliance Complexity: Each plan required comprehensive checks against ESSA (Every Student Succeeds Act) federal regulations and multiple educational frameworks, demanding deep subject matter expertise and attention to detail.
- Resource Allocation Issues: The labor-intensive review process diverted valuable administrator time away from focusing on student outcomes and strategic educational initiatives.
- Scalability Constraints: Traditional manual approaches could not scale to meet growing demand from school districts requiring rapid, accurate feedback on their improvement plans.
- Quality Consistency Concerns: Manual reviews risked inconsistencies in feedback quality and compliance verification across different reviewers and time periods, potentially exposing schools to regulatory issues.
Key Results
- Reduced Review Time by 80%: Automated the complete review process from 60-100 page school improvement plans, delivering comprehensive compliance reports in 10-15 minutes versus hours or days of manual work.
- Achieved Comprehensive Compliance Coverage: Implemented multi-faceted analysis covering ESSA federal regulations, SMART/BEST goal frameworks, coherence checks, and educational best practices validation.
- Enabled Scalable Operations: Deployed a serverless architecture capable of processing multiple school plans simultaneously without additional staffing requirements.
Solution
Event-Driven Serverless Architecture: Designed and implemented a fully automated, serverless solution on AWS using parallel Lambda functions triggered by S3 file uploads, enabling immediate processing upon plan submission through a web-based user interface.
Advanced Document Processing Pipeline: Developed a sophisticated multi-stage analysis system that performed section-wise content extraction from PDF documents, including tabular data and logo extraction, followed by automated coherence analysis validating logical connections between problem statements, root causes, and strategic objectives.
AI-Powered Compliance Verification: Integrated Amazon Bedrock with Claude models to conduct intelligent compliance checks against federal regulations, leveraging a custom AWS Vector Knowledge Base containing ESSA documents, licensed planning guides, and educational best practice resources.
Intelligent Document Chunking Strategy: Implemented path-based chunking mechanisms for knowledge base ingestion, applying specialized strategies for different document types—section-wise chunking for ESSA regulations, whole-document treatment for reference materials, and fixed-size chunking with overlap for generic documents.
Retrieval-Augmented Generation (RAG) System: Built a comprehensive RAG pipeline that queried the vector knowledge base to validate goals, needs assessments, and strategic plans against regulatory requirements, ensuring accurate compliance reporting.
Structured Report Generation: Engineered a generative AI system that synthesized multiple analysis outputs (parsing, coherence, goal validation, and compliance checks) into structured “Glows and Grows” format reports with executive summaries and section-by-section feedback.
Real-Time Status Monitoring: Implemented a polling mechanism with 20-second intervals to track report generation progress, delivering completed reports to an editable markdown interface for final user review and customization.
Production Deployment Infrastructure: Deployed the Next.js user interface on AWS EC2 (t2.small instance) with proper security configurations, enabling external access and providing screen session management for continuous application availability.
Technologies Used
- AWS Lambda
- Amazon Bedrock (Claude Models)
- AWS S3
- Amazon Knowledge Base for Bedrock
- AWS Vector Database
- PyMuPDF (fitz)
- Python (boto3)
- Next.js
- AWS EC2