Build a Custom Voice AI for Automated Reference Checks
The best voice AI reference checking solution is a custom system using a large language model. It automates phone calls, transcribes responses, and flags risks based on your criteria.
We built a system for a 12-person recruiting firm that was processing 400 applicants per month. Their recruiters spent 30 minutes per candidate on manual calls. The new system completes five reference checks in under 10 minutes of automated call time and delivers structured summaries directly to their ATS, cutting recruiter time spent on this task by 90%.
A custom build is ideal for small businesses that need consistent, unbiased reference data without per-seat SaaS fees. The system is built on your infrastructure, integrates directly with your Applicant Tracking System (ATS), and uses your specific reference questions. This approach is for teams who see reference checks as a critical data source, not just a checkbox.
What Problem Does This Solve?
Recruiting teams often start with manual phone calls for reference checks. This process is slow and inconsistent. One recruiter might ask probing follow-up questions while another sticks to the script, making it impossible to compare candidates objectively. A single senior candidate requiring five reference calls can consume half a day of a recruiter's time.
Trying to scale this, teams turn to survey tools like Typeform or built-in ATS features. These just send an email with a link to a form. Response rates are often below 30% because it puts the work on the reference. A reference who would happily talk for 10 minutes will ignore a form that looks like a 20-minute task. There is no way to ask dynamic follow-up questions based on an initial answer.
Dedicated reference checking platforms like Checkster exist, but they charge per-candidate or per-seat, which is expensive for a small business. A 5-person recruiting team can face a bill of over $500/month. These platforms also offer limited customization of the analysis logic, flagging generic risks instead of the specific competencies you screen for.
How Does It Work?
Our process starts by codifying your existing reference questions and decision criteria into a structured script. We use Anthropic's Claude 3 Sonnet API to manage the conversational flow, allowing for dynamic follow-up questions that dig deeper into a reference's initial response. This logic is deployed as a Python service using FastAPI.
When a recruiter triggers a check from your ATS, a webhook fires an AWS Lambda function. This function uses the Twilio API to place the outbound call. Audio is streamed in real-time to AWS Transcribe for speech-to-text conversion. The transcribed text is sent to the Claude API, which generates the next question. The response is then converted back to audio using AWS Polly and played to the reference. The entire question-response loop takes under 800ms.
All call data, including full audio recordings, transcripts, and a final structured summary, is written to a Supabase database. The summary, which includes flags for hesitation, negative sentiment, or conflicting information, is posted back to a custom field in your ATS via its API. A recruiter sees a complete report within 60 seconds of the final call ending. The total cloud services cost for 200 five-minute reference calls is typically under $40 per month.
We build a simple dashboard showing call completion rates, average call duration, and common keywords from transcripts. The system includes structured logging with structlog, sending alerts to Slack via webhooks if a call fails due to a technical issue or if transcript quality scores are low, indicating a bad connection. This allows for immediate review without disrupting the entire process.
What Are the Key Benefits?
From Kickoff to First Automated Call in 3 Weeks
Your custom system is live and integrated with your ATS in 15 business days, not a full quarter. Start getting structured reference data immediately.
A Fixed-Price Build, Not a Recurring Subscription
You pay a one-time project fee. After launch, you only pay for cloud usage, which is often less than $50/month, with no per-seat or per-check fees.
You Own the Source Code and the Data
We deliver the complete Python codebase to your GitHub repository. Your data resides in your own database, not on a third-party vendor's platform.
Real-Time Alerts for Failed Calls
A health check runs every 5 minutes. If API connections fail or calls do not complete, you get an immediate Slack notification with the error details.
Connects Natively With Your ATS
We use direct API integrations to pull candidate data from and push reports to Greenhouse, Lever, or any ATS with an accessible API. No manual data entry is needed.
What Does the Process Look Like?
Discovery and Scripting (Week 1)
You provide your reference questions, ATS API credentials, and examples of good and bad reference checks. We deliver a detailed technical plan and a final conversational script.
Core System Build (Week 2)
We build the FastAPI service, Lambda functions, and Supabase schema. You receive a private link to a staging version to test internal calls and review transcript accuracy.
Integration and Deployment (Week 3)
We connect the system to your ATS and deploy it to your AWS account. You receive credentials, and we process the first batch of 10-20 real candidate references together.
Monitoring and Handoff (Weeks 4-6)
We monitor system performance and data quality for two weeks post-launch. You receive a final runbook detailing the architecture, monitoring dashboards, and maintenance steps.
Frequently Asked Questions
- How much does a custom voice AI reference system cost?
- The cost is a fixed-price project fee based on scope. Key factors include the number of unique question paths, the complexity of the risk-flagging logic, and the specific ATS integration. It's a one-time build, so there are no recurring license fees. We can provide a detailed quote after a 30-minute discovery call where we review your exact requirements.
- What happens if a reference hangs up or gives a short answer?
- The system logs the call duration and outcome. If a call is shorter than a preset threshold, like 90 seconds, it's flagged for manual review. If a reference hangs up, the system logs the incomplete status and notifies the recruiter. The logic can be configured to detect terse answers and ask a clarifying follow-up question before ending the call.
- How is this different from using a service like Checkster?
- Checkster is a multi-tenant SaaS platform where you are one of thousands of customers. Syntora builds a single-tenant system that you own and control completely. You are not subject to their feature roadmap or pricing changes. Most importantly, you own the underlying code and can modify the analysis and scoring logic anytime without waiting for a vendor.
- Do we need to get consent from references to record calls?
- Yes. The system's introductory message explicitly states that the call is an automated reference check and is being recorded. It then asks the reference to give verbal consent to continue. If the reference says no or does not respond, the call is terminated and logged accordingly. This consent step is a mandatory part of every call flow we build.
- Can this system handle different languages?
- Yes, but it's scoped per language. The underlying services we use, AWS Transcribe and Polly, support dozens of languages. A build for Spanish would require a separate conversational script and would be a distinct part of the project scope. We typically build for one primary language initially and can add others later as needed. Each language adds about one week to the build time.
- What technical skills are needed to maintain this system?
- Basic maintenance, like changing a question in the script, is done by editing a text file. For deeper changes, a developer comfortable with Python and REST APIs would be required. The system is delivered with a runbook that explains the architecture and common maintenance tasks. For teams with no technical staff, we offer a flat monthly maintenance plan to handle updates and monitoring.
Related Solutions
Ready to Automate Your Small Business Operations?
Book a call to discuss how we can implement ai automation for your small business business.
Book a Call