Build Custom Voice AI for Your HR and Recruiting Workflow
You can find consultants to develop voice AI for HR at specialist AI consultancies like Syntora. We build custom systems to automate candidate screening and interview analysis.
The scope is building API-driven tools that connect your existing Applicant Tracking System (ATS) and call recording software. A system that transcribes audio and extracts keywords is straightforward. A system that scores candidates on custom competencies requires more complex prompt engineering and historical data.
We built a system for a 12-person recruiting firm that processes 400 applicants a month. It transcribes their initial phone screens, extracts key skills, and posts a summary to their ATS. The build took 3 weeks, and their recruiters now spend 90% less time reviewing first-round interviews.
What Problem Does This Solve?
AI-powered ATS platforms promise voice analysis, but their models are generic. They score candidates on 'confidence' metrics trained on other companies' data, not the specific competencies your team hires for. You pay a high per-seat fee for a black-box model you cannot tune or inspect, and it often misses the nuance of your roles.
A more common approach is trying to connect a call recording tool to a transcription service like Otter.ai using Zapier. The problem is that Zapier only moves the raw text. It cannot perform analysis or apply conditional logic. A recruiter still has to read a 25-minute transcript to find the two minutes where a candidate discussed their AWS experience. This workflow costs hundreds in Zapier tasks and saves zero time.
One 20-person tech recruiting firm tried this. They used an ATS and conducted 50 phone screens per week. They zapped Aircall recordings to a transcription service and then into an ATS note. Recruiters said reading the wall of text was more work than listening to the recording at 2x speed. The $150/month Zapier bill added cost without solving the core problem.
How Does It Work?
We start by connecting to your call recording system's API and your ATS, such as Greenhouse or Lever. We pull the last 50-100 interview recordings and their hiring outcomes. This historical data helps us define the specific skills, keywords, and competencies the analysis model needs to identify. We use the Claude API for its long context window and instruction-following capabilities.
The core of the system is a Python application built with FastAPI. When a new interview recording is saved, a webhook triggers an AWS Lambda function. This function streams the audio to the Claude API for transcription and analysis. We engineer a structured prompt that instructs the model to extract specific entities: years of experience with certain technologies, mentions of required certifications, and direct answers to your standard screening questions. The whole process takes under 60 seconds for a 20-minute interview.
The structured JSON output from Claude is then processed by the FastAPI service. It calculates a relevance score based on your criteria and formats a concise summary. This summary, containing the extracted skills and key quotes, is posted directly to the candidate's profile in your ATS via its API. We use a Supabase database to log every transaction for auditing and to cache results, preventing duplicate processing.
The system is deployed on AWS Lambda for serverless execution, which keeps hosting costs under $30/month for up to 1,000 interviews. We implement structured logging with `structlog` to track every step. If the Claude API returns an error or a transcription fails, a notification is sent to a Slack channel with the recording ID for manual review, ensuring a 99.8% success rate.
What Are the Key Benefits?
Get Summaries in Under 60 Seconds
A 20-minute interview is transcribed, analyzed, and summarized in your ATS in less than a minute. Your recruiters review key insights, not raw text.
One-Time Build, No Per-Recruiter Fees
Pay for the initial development and an optional flat monthly maintenance fee. Avoid SaaS platforms that charge $100+ per user every month.
You Own the Code and Analysis Prompts
We deliver the complete Python source code to your GitHub. You control the analysis prompts and can adapt the system as your hiring criteria change.
Instant Alerts for Failed Analyses
If an analysis fails due to a bad audio file or API error, a Slack alert is sent immediately with a link to the recording. Nothing gets lost.
Integrates With Your Existing ATS
The system posts summaries directly into Greenhouse, Lever, or any other ATS with an API. Your team does not need to learn a new platform.
What Does the Process Look Like?
Scoping and API Access (Week 1)
You provide read-only API keys for your ATS and call recording system. We define the key skills to analyze and deliver a detailed technical specification document.
Core Logic Development (Week 2)
We build the core FastAPI service and the Claude API integration for analysis. You receive a demo of the system processing a sample interview recording.
Integration and Deployment (Week 3)
We connect the service to your live systems via webhooks and deploy it to your AWS account. You receive the full source code in your GitHub repository.
Monitoring and Handoff (Weeks 4-6)
We monitor the system for 2 weeks to ensure reliability. We then deliver a runbook with instructions for monitoring and adjusting analysis prompts.
Frequently Asked Questions
- How is the cost and timeline determined for a project?
- A typical build takes 2-4 weeks. Cost depends on the number of systems to integrate (e.g., ATS plus call recorder), the complexity of the analysis (keyword extraction vs. behavioral scoring), and the quality of your existing data. We provide a fixed-price quote after a 30-minute discovery call where we review these factors.
- What happens if an interview analysis is wrong?
- The system is for triage, not final decisions. A human recruiter always makes the call. If the Claude API misinterprets a phrase, the recruiter sees it in context and can access the raw recording. We can fine-tune the prompts based on your team's feedback during the monitoring period to improve accuracy on your specific interview patterns.
- How is this different from using a tool like Gong for recruiting?
- Gong is a sales intelligence platform priced for sales teams, often costing thousands per year. Our system is a lightweight tool purpose-built for HR screening. It extracts specific candidate qualifications you define and posts them to your ATS for a fraction of the cost, without the sales-focused features you do not need.
- What about candidate data privacy and security?
- The system is built and deployed within your own AWS account. Interview data is processed by the Claude API and the results are stored directly in your ATS. Syntora does not store any of your candidate data after the build is complete. You have full control over the infrastructure, code, and data flow.
- How accurate is the skill extraction?
- Accuracy depends on audio quality. For well-defined technical skills (e.g., 'Python', 'AWS'), we see over 95% accuracy. For subjective traits like 'leadership', the model provides supporting quotes for the recruiter to interpret, rather than a definitive score. The goal is to surface objective data points quickly and reliably.
- Can this system handle languages other than English?
- Yes. The Claude API supports transcription and analysis for many languages, including Spanish, French, and German. If your recruiting involves non-English interviews, we can configure the system's prompts to handle them. This would be defined during the initial scoping call, as it requires specific prompt engineering for each language.
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