Overview
Upward Physical Therapy needed practical automation that could save time without adding enterprise-level software overhead. The opportunity was to use AI agents to support two high-friction workflows: recruiting licensed PT candidates and answering financial growth questions like whether the clinic could afford another hire or expanded hours.
Problem
Like many small clinics, Upward PT was managing important operational work manually. Candidate screening required reviewing resumes and tracking fit across multiple criteria, while financial planning relied on exported reports and ad hoc analysis.
The clinic needed a lightweight system that could reduce manual effort, surface better decisions faster, and keep sensitive data in a controlled environment. The broader tech review also identified missing capabilities in recruiting and financial analytics within the clinic's existing tool stack.
Solution
Recruiter Bot
- Monitors a recruiting inbox
- Downloads PDF and DOCX resumes
- Extracts and normalizes candidate data
- Scores applicants using a deterministic rubric
- Tracks candidate status in a pipeline
- Sends daily digests and draft response emails for review
CFO Bot
- Ingests exported QuickBooks CSVs
- Computes clinic KPIs
- Models scenarios like hiring another PT, expanding hours, or adding a treatment room
- Answers natural-language financial questions
- Produces monthly summary reports
The system was intentionally designed with human approval gates, local storage, audit logging, and security mitigations for prompt injection and sensitive data handling.
Key Capabilities
- AI-assisted resume intake and scoring
- Daily candidate digest delivery to mobile chat
- Automated KPI computation from accounting exports
- Natural-language scenario analysis for hiring and expansion decisions
- Local-first architecture with controlled outbound actions
Technical Approach
- Claude multi-agent runtime
- Matrix / Element notifications
- IMAP email ingestion
- PDF and DOCX extraction
- SQLite or Google Sheets for pipeline tracking
- CSV ingestion and deterministic KPI calculations
- Claude-powered Q and A and summarization