Automate Income Anticipation for Affordable Housing

Property management platforms like RealPage and AppFolio have application portals but lack sophisticated income calculation logic. Their systems require leasing agents to manually read pay stubs and enter data. This works for simple W-2 income but fails with variable sources like tips, gig work, or seasonal overtime, creating a major compliance risk.

A leasing team for a new 500-unit LIHTC property receives 1,500 applications. An applicant has two jobs: a part-time hourly role and freelance design work documented with bank statements. The leasing agent must manually calculate the average hourly rate, project it over 2080 hours, then separately average the last three months of deposits from the bank statement. This 25-minute manual process, repeated 1,500 times, creates a 625-hour backlog, delaying lease-up by months.

This manual approach is not just slow; it is inconsistent. Different agents may interpret the same documents differently, leading to applicants being incorrectly sorted into AMI buckets. An error can result in a rejected application for a qualified household or a compliance failure discovered during a state audit, jeopardizing the property's tax credits.

Syntora would approach this problem by first conducting a detailed discovery phase to understand your specific program rules, existing property management software integrations (RealPage, AppFolio), and the variety of income documents processed. This foundational understanding would inform the bespoke system architecture.

At its core, the system would involve a custom API service, likely built with FastAPI, designed to integrate directly with your existing application submission workflow via webhooks. Upon application submission, an event would trigger an AWS Lambda function to retrieve application data and any uploaded income documents. For document processing, we would leverage large language models such as the Claude API. Our team has experience building robust document processing pipelines using the Claude API for complex financial documents, and a similar pattern would apply here to reliably extract critical fields like employer, pay rate, hours worked, and year-to-date earnings from various pay stub formats.

The extracted data would then feed into a custom Python calculation engine. This engine would be designed with modular logic to accurately annualize various income types including hourly wages, tips, commissions, and one-time bonuses. All calculations would be tailored precisely to your specific LIHTC, HOME, or other program requirements, including any necessary asset verification triggers.

The calculated 12-month anticipated income would be compared against your property's AMI tiers, which would be managed in a scalable database like Supabase. The system would classify the applicant into the appropriate AMI bucket (e.g., 50% AMI, 60% AMI). This result, along with a link to a detailed parsed data summary, would then be written back to a custom field within your RealPage or AppFolio instance, automating a significant portion of the eligibility screening.

For system observability, we would integrate structured logging (e.g., with `structlog`) and CloudWatch alarms to monitor performance and identify processing anomalies. If the Claude API's confidence in parsing a document falls below a defined threshold, an alert would be routed for manual review, ensuring accuracy and compliance. A typical build timeline for a system of this complexity, from discovery to deployment, would range from 10 to 16 weeks, depending on the intricacies of integration and rule sets. The client would typically provide access to APIs, sample documents, and clear definitions of income calculation rules and AMI tiers. Deliverables would include the deployed system, source code, documentation, and training for your team.