jaysahnan/gro-982-extend-browserbase-template #56
Conversation
shrey150
left a comment
shrey150
left a comment
There was a problem hiding this comment.
Overall the template works, but there's a significant amount of unnecessary complexity that should be cleaned up before merging — especially since this is a template people will clone and learn from.
The biggest finding: both extend-ai and @browserbasehq/sdk already have built-in retry with exponential backoff (default 2 retries for 408/429/5xx). The hand-rolled retry logic throughout this PR is redundant and should be removed in favor of the SDK defaults (or bumping maxRetries if needed). Both SDKs also expose typed error classes (ExtendError and Browserbase.APIError) — use those instead of string-matching on error messages.
| print("Session not found, returning empty result") | ||
| return 0 | ||
| # HTML error response - session may not be ready yet, keep retrying | ||
| if "Unexpected token '<'" in error_message or "<html" in error_message: |
There was a problem hiding this comment.
Same as the TS version — @browserbasehq/sdk likely exposes typed errors. The string matching on "Session with given id not found", "-32001", "Unexpected token '<'", and "<html" is brittle. Check if the Python SDK has an equivalent of Browserbase.APIError with a status code property.
| } | ||
|
|
||
| // Process in batches of 9 to balance speed and reliability | ||
| for (let i = 0; i < filePaths.length; i += 9) { |
There was a problem hiding this comment.
Why 9? If this is an Extend rate limit, document it. If arbitrary, extract to a named constant with a comment explaining the choice.
There was a problem hiding this comment.
[shrey] +1 on avoiding a magic number here
shrey150
left a comment
shrey150
left a comment
There was a problem hiding this comment.
Pre-approving, consider pinning version numbers to avoid extra engineering work
| @@ -12,7 +12,8 @@ | |||
There was a problem hiding this comment.
Sure we don't want to a pin a version number here? e.g. if we set "stagehand": "^3.0.0" so if for example we release v4 w/ breaking changes, this script remains functional
2 participants
Summary
Adds a new Extend + Browserbase template that automates downloading expense receipts from a web portal and parsing them with Extend AI for structured data extraction (vendor, date, totals, line items, payment method).
Includes both TypeScript (typescript/extend-browserbase/) and Python (python/extend-browserbase/) implementations with identical functionality.
Uses the observe -> act pattern to find and click all individual download buttons, polls Browserbase's Downloads API for the session ZIP, extracts files, then uploads to Extend for schema-based receipt extraction. Results are saved as JSON and CSV.