Build Your Own Voice AI Recruiting and Scheduling System

Startups first try to solve this with scheduling links from tools like Calendly. This works for booking calls but does not screen candidates. Recruiters waste hours on calls with applicants who don't meet basic requirements, like work authorization or specific years of experience, because Calendly cannot ask these knockout questions first.

Next, they look at dedicated AI recruiting platforms like Paradox or Humanly. These platforms are powerful but built for large enterprises. They often require annual contracts starting at five figures and have minimums of 500+ employees. For a 30-person startup, paying for features like advanced compliance reporting and multi-language support is unnecessary overhead. The sales process is long and the implementation takes a full quarter.

A common failure scenario involves a recruiting agency trying to use a patchwork of tools. They use their ATS to track candidates, then manually export a list to an email marketing tool to send screening questions, then manually update the ATS based on replies. This process is slow, error-prone, and a data-syncing nightmare. One client was losing 10% of qualified candidates who dropped off during the multi-day lag between application and first contact.

Syntora's approach to building a voice AI recruiting and scheduling system begins with a discovery phase to understand your specific hiring workflows and screening criteria. The initial step involves establishing secure API connections to your Applicant Tracking System, whether it is Greenhouse, Lever, or Ashby. The system would be designed to utilize webhooks to detect new candidate applications in real-time.

For each new candidate, their resume and application data would be processed. We have experience building document processing pipelines using Claude API for sensitive financial documents. Applying this same pattern, the system would parse candidate information with custom prompts, extracting structured data points such as skills, years of experience, and previous employers. Our experience indicates this extraction can be very efficient, often completed within a few seconds.

Based on your predefined rules, the system would evaluate if a candidate is a potential fit. If criteria are met, a Python script running on AWS Lambda would trigger an SMS via Twilio to initiate pre-screening questions. For example, for a senior engineer role, this might involve a direct question like 'Do you have 5+ years of production Python experience? Yes/No'. The candidate's response would be processed, and if they pass, a unique scheduling link would be sent. Every interaction and decision point would be logged back to the ATS as a note on the candidate's profile.

The core logic of the system would be implemented as a FastAPI service. Event logging would be managed using structlog, and conversation history would be stored in a Supabase Postgres database. This architecture enables detailed reporting on candidate drop-off points within the screening process. We estimate the infrastructure cost for processing 1,000 candidates per month would typically be under $50, covering AWS Lambda, Supabase, and Twilio usage.

As a key deliverable, Syntora would provide a simple dashboard to monitor the system's performance. This dashboard would track metrics like response rates to initial outreach, pass-through rates for each screening question, and the number of interviews successfully booked. To ensure reliability, if an API call to your ATS fails more than three times, a structured log alert would be sent to a dedicated Slack channel for immediate investigation and resolution.