Automate BPO Insurance Claims Intake with AI SDK and Cognition Devin Agent
Offshore BPO processes require high-precision automated intake workflows to prevent claims processing errors. Build a streaming claims intake application using Vercel AI SDK and simulated Cognition Devin coding agent automation. Achieve under 2 seconds end-to-end processing latency per claim with 95% automated extraction precision.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a TypeScript/Node.js claims processing microservice using Vercel AI SDK integrated with Cognition Devin task workflows.
How work is evaluated
Evaluates TypeScript AI SDK streaming pipeline speed and field extraction accuracy.
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.
latency_check
Total processing time per document is less than 2000 milliseconds.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
field_extraction_f1
F1 score across extracted structured fields • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Implement Vercel AI SDK `generateObject` and `streamObject` methods.
Integrate Cognition Devin capabilities for code generation and workflow setup.
Build type-safe claims schema extraction using Zod in TypeScript.
Measure stream latency and schema validation accuracy.
Reference links and supporting material
Dataset of 300 unstructured customer call logs and incoming emails for insurance claims BPO intake.
Challenge setup
Evaluates TypeScript AI SDK streaming pipeline speed and field extraction accuracy.
Challenge input
Unstructured text string representing intake call report
AI SDK
Vercel standard framework for type-safe TypeScript AI apps.
Cognition Devin
Automates setup, testing, and continuous deployment workflows.
Evaluated output
Validated Zod JSON object with caller_name, loss_date, loss_type, estimate
- Total processing time per document is less than 2000 milliseconds.
- F1 score across extracted structured fields • target: 0.95 • range: 0-1
- Benchmark: StructuredExtractionBench
- Field Extraction F1 target: 0.95
- 1 public reference case
- JavaScript execution harness
- javascript sandbox (unavailable on Versalist)
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
- Cognition Devin · Required
- Aim · Optional
- Seldon Core · Optional
- Cognition Devin · Required
- Cognition Devin · Required
Evaluation contract
- latency_check · Weight 1
- field_extraction_f1 · Weight 1
Metrics
- field_extraction_f1 · Target 0.95
Benchmarks
- StructuredExtractionBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: Unstructured text string representing intake call report
Output: Validated Zod JSON object with caller_name, loss_date, loss_type, estimate
Execution harness: JavaScript
Evaluation tools
- AI SDK · Agent Framework
- Cognition Devin · Autonomous Developer Agent
Recipe freeze marker
No recipe freeze time is recorded.