Automate FOIA Request Processing
This challenge requires designing and implementing an intelligent automation system to efficiently process and filter high volumes of incoming Freedom of Information Act (FOIA) requests. Leveraging Vercel's AI SDK, participants will build a responsive web interface for request submission and monitoring. OpenAI o4-mini will serve as the core AI engine for document classification, summarization, and anomaly detection to identify spam or 'robo-requests'. Key aspects include integrating Unstructured.io for robust document parsing, orchestrating the multi-step workflow with Lindy for compliance and auditing, and setting up Langsmith for comprehensive observability of AI agent decisions and human review queues. The solution aims to streamline administrative tasks, reduce human workload, and ensure that legitimate FOIA requests are processed accurately and efficiently.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an AI SDK application with OpenAI o4-mini and Lindy to automate the processing, classification, and routing of high-volume FOIA requests.
How work is evaluated
Evaluate the system's accuracy in classifying FOIA requests, efficiency in processing, and effectiveness in detecting and filtering 'robo-requests'.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
spam_detection_rate
Checks if a predefined set of known spam/robo-requests are correctly identified and filtered out.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
legitimate_request_routing
Verifies if legitimate requests are correctly classified and routed to the appropriate next step in the Lindy workflow.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
output_format_compliance
Ensures that extracted information adheres to the expected structured format (e.g., specific JSON schema).
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
classification_accuracy
Percentage of FOIA requests correctly classified into their respective categories. • target: 0.95 • range: 0.85-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
processing_time_per_request_sec
Average time (in seconds) taken to process a single FOIA request from input to classification. • target: 3 • range: 1-10
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
data_extraction_f1_score
F1 score for accurately extracting all key information fields from legitimate requests. • target: 0.9 • range: 0.8-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master building AI-powered web applications using Vercel's AI SDK, focusing on streaming responses and tool integrations for a responsive user experience.
Implement document parsing and information extraction using Unstructured.io to process various FOIA request formats (PDFs, emails, text files).
Leverage OpenAI o4-mini for intelligent classification of incoming requests (e.g., legitimate, spam, ambiguous) and for summarizing key request details.
Design and orchestrate an automated FOIA request processing workflow using Lindy, integrating AI classification steps, human review queues, and external system updates.
Develop an anomaly detection module using OpenAI o4-mini to flag suspicious or repetitive 'robo-requests' based on patterns in the request text and metadata.
Integrate Langsmith for comprehensive observability, tracing AI agent decisions, tool calls, and human review steps throughout the FOIA processing pipeline for auditability.
Build a user-friendly dashboard using the AI SDK to monitor the status of requests, review classified documents, and manage human intervention points.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
Evaluate the system's accuracy in classifying FOIA requests, efficiency in processing, and effectiveness in detecting and filtering 'robo-requests'.
Challenge input
Array of JSON objects, each with 'request_id' (string) and 'content' (string) fields.
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON object where keys are 'request_id's and values are predicted classification categories (string).
- Checks if a predefined set of known spam/robo-requests are correctly identified and filtered out.
- Verifies if legitimate requests are correctly classified and routed to the appropriate next step in the Lin...
- Ensures that extracted information adheres to the expected structured format (e.g., specific JSON schema).
- Classification Accuracy target: 0.95
- 2 public reference cases
- JavaScript execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- spam_detection_rate · Weight 1
- legitimate_request_routing · Weight 1
- output_format_compliance · Weight 1
- classification_accuracy · Weight 1
- processing_time_per_request_sec · Weight 1
- data_extraction_f1_score · Weight 1
Metrics
- classification_accuracy · Target 0.95
- processing_time_per_request_sec · Target 3
- data_extraction_f1_score · Target 0.9
Reference cases
- 2 public reference cases
Workflow tasks
- Task 1
Input: Array of JSON objects, each with 'request_id' (string) and 'content' (string) fields.
Output: JSON object where keys are 'request_id's and values are predicted classification categories (string).
Execution harness: JavaScript
- Task 2
Input: JSON object with 'request_id' (string) and 'content' (string) for a legitimate request.
Output: JSON object with extracted fields: 'requesting_party' (string), 'subject_matter' (string), 'date_range' (string).
Execution harness: JavaScript
Recipe freeze marker
No recipe freeze time is recorded.