phData

Get the latest jobs and industry news

$1M+ in annual savings, and new AI use cases ship 5× faster. One of the largest K-12 edtech companies in the US spent $1.5M and 12+ months with a prior vendor, and the AI tutor hallucinated in front of the CEO. phData delivered a working proof of concept in 10 weeks using Claude on Amazon Bedrock and Snowflake. Every AI system now passes tests against synthetic students before it reaches a student. See how it came together.

Six to eight hours. Every day. That’s how long a $9B+ cruise line spent closing out employee incentive calculations on a decade-old Oracle and MS Access system nobody could safely change. phData rebuilt it on Snowflake in about six to seven months. Daily batch time dropped to three to four hours. Output matched the legacy system within 1%. All three brands now run on one platform, and business users handle split commissions and other exceptions in a Streamlit app. See how the team turned a decade of undocumented rules into testable dbt models.

$1.4M invested. $4.1M saved. A $1B+ healthcare services company was processing 320,000 payer dispute emails a month through 1 onshore employee and 100 offshore staff, with $290M in receivables tied to No Surprises Act arbitration deadlines. phData built a platform with Claude on Snowflake Cortex in 4 months. Default losses and administrative fees dropped 50%, and every email is now classified automatically at 90%+ accuracy. See how it came together.

3 days to answer a contract question. Under 5 seconds with Claude. A $7.4B financial services firm was fielding 250 contract inquiries a month with 5 employees, and volume was growing 40% a year. phData built a RAG assistant on Amazon Bedrock that drafts each answer and cites the exact contract clause behind it. Customer contract data stays inside the firm’s AWS environment, and the team absorbs the inquiry growth without adding headcount. See how it came together.

Vendor orders that took staff hours now finish in minutes, with a 100% success rate across vendors. Order.co processes nearly $1B in annual purchasing volume, and vendor websites change constantly, which breaks scripts built on fixed click sequences. phData built automation in 6 weeks that reads a page and reasons through changes, using Claude on Amazon Bedrock. Two QA agents verify the cart and checkout totals before any order is submitted, and structured error codes tell operations exactly what failed. Here’s how the build worked.