Industry Specific, Responsible AI
Generic note-taking and summarization solutions fall short when it comes to truly saving time and helping make important staffing decisions.
HireLogic’s team of data scientists built proprietary machine learning models trained on millions of interview minutes, resumes, and job descriptions, to understand the context of hiring conversations to extract valuable insights that go beyond notes to help humans make better decisions.
At the heart of decision support lies the challenge of translating data into actionable insights. HireLogic tackles this ‘last mile’ problem head-on, with a unique combination of fine-tuned models coupled with best-in-class commercial models to analyze your conversations and deliver the most accurate and objective insights possible.
The HireLogic team is continuously monitoring AI regulations to ensure your use of the solution is in compliance with such laws. Furthermore, HireLogic is the only HR-focused organization that participates in the development of responsible AI standards, such as ISO/IEC 42001.
Enterprise Security, Privacy, and Scalability
HireLogic is architected following modern application security best practices, including consideration for OWASP top ten web application security risks, and is deployed on secure, SOC-certified cloud infrastructure used by other large enterprises and Applicant Tracking Systems (ATS). All data, including candidate information, is encrypted both in transit and at rest, in accordance with industry standards and best practices.
HireLogic uses a robust microservices architecture that dynamically scales concurrent interview processing tasks, supporting some of the largest product deployments of interview intelligence.
HireLogic follows a conservative approach to data retention, minimizing the storage of audio or video recordings, offering zero-transcript notes, automatically redacting personal data (PII), and offering custom data retention policies.
Process-Driven Integration
Based on real world usage across millions of interview minutes, HireLogic has developed a streamlined user experience and integrated with a wide variety of Applicant Tracking Systems (ATS), Telephony Providers, Video Conference Platforms, and Directory Services, to fit seamlessly into your current technology stack.
From capturing conversations in any online or offline mode, to sending valuable insights directly into your ATS, HireLogic’s integration strategy and custom services offerings ensures that we are able to minimize onboarding friction and training requirements for busy recruiters.
HireLogic supports a variety of enterprise Single Sign-on (SSO) options, including Okta and Active Directory (SAML and OIDC), for easy deployment, and even offers a “zero login” option where insights are automatically captured and flow directly into your ATS.
HireLogic Contextual AI Pipeline
Proprietary hiring-focused models, not just a LLM pass-through
1. Transcription and Speaker Diarization
HireLogic converts audio streams into transcripts using industry-leading speech to text services. We do not persist audio or video, preferring the lowest footprint option to minimize bias, maximize privacy, and reduce the need to review recordings.
Integrate to audio stream or file depending on interview mode
Retrieve audio stream from any telephony-based, video, or in-person (device mic) conversation.
Transcribe the conversation
Use best-in-class transcription services based on continuous evaluation of services across word error rate, latency, and other factors.
Detect individual speakers
Use best-in-class speaker diarization services to determine individual speaker segments (used in later step to determine role).
Personal info (PII) redaction
Automatically redact personal information such as social security numbers, credit card numbers and other sensitive information.
2. Extraction and Pre-processing
Using proprietary algorithms and fine-tuned AI models, HireLogic detects the role of each speaker in a multi-party conversation, filters out typical early casual conversation unrelated to the job, and extracts relevant conversation segments for abstraction by leading LLMs.
Translation if needed
Auto-detect over 20 languages to analyze context and provide insights in that local language.
Detect speaker roles
Proprietary, fine-tuned models to determine the speaker roles across candidate and interviewer(s).
Small talk filtering
Proprietary, fine-tuned models to filter out casual conversation topics unrelated to job discussion, typically at the start of a conversation.
Parse and extract hiring-specific items
Detect and extract hiring-specific and job relevant information from the conversation, to highlight important elements.
Prepare data for custom chatbot
Segment the conversation into appropriate sized chunks for vectorization and RAG training.
3. Abstraction
Using a combination of proprietary and commercial language models, HireLogic abstracts and summarizes per-filtered conversational elements into insights that can be used to inform human decision making.
Abstract and summarize the pre-processed data
Leverage multiple large language models (LLMs) to summarize specific segments of the conversation to make them easier to consume.
“Noise” reduction from transcription error
Filter out noise or redundant information that could be the result of transcription errors.
Proprietary dynamic prompts for purpose-built insights
Utilize dynamic prompts and proprietary algorithms to extract specialize insights and discrete elements for candidate evaluation.
4. Post-Processing
Even the best commercially available LLMs are far from perfect and will never be fully trained on organic interviews. In this step, HireLogic uses proprietary models and algorithms to filter out hallucinations, contradictions, and subjective statements, to increase accuracy and consistency. This is critical to trustworthy, responsible data.
Parse and organize data
Organize the abstracted data into various categories for further processing.
Filter irrelevant & subjective outputs
Apply proprietary algorithms and thresholds to detect and filter any subjective abstractions or statements.
Filter contradictions, and hallucinations
Apply proprietary algorithms and thresholds to detect and filter any potential contradictory statements or hallucinations.
Leverage multiple LLMs to improve accuracy and reduce “laziness”
Leverage multiple large language models (LLMs) to further refine insights and improve accuracy.
Back-mapping and/or translation if needed
Apply back-mapping to translate insights if needed.
5. Presentation
Finally, the insights are formatted and presented in a digestible, easy to understand format, across any destination: email, your ATS or HCM system, or within HireLogic. An easy to use, private chatbot is trained to answer any question about any candidate or conversation that you invited HireLogic to, for deeper insights that may not be available in the default reports.
Present data in a way that is easy for recruiters and hiring managers to understand
Insights are consistently formatted in a way that is easy to understand and compare across candidate.
Visually segment information around candidate and interview
Insights are organized and presented for easy consumption based on extensive user research and customer feedback.
Enable custom chatbot queries via RAG training
Provide an easy to use chatbot interface to answer any questions and provide further insights beyond the default scope.
