Automating Interview Reporting with LLM-Powered Transcript Analysis

Quick Summary

Challenge

Manual interview summaries were inconsistent and time-consuming.

Solution

Tatras Data developed a multi-stage pipeline to summarize interviews, detect sentiment, and map candidate responses to job criteria in under 3 minutes.

Result

Reduced interview report turnaround time by over 90%.

Tech Stack

AI: OpenAI + open-source LLMs | ML: Sentiment analysis Summarization Gap detection | Data & Retrieval: MongoDB Semantic mapping | Dev: AWS Step Functions Lambda orchestration Scalable microservices | Viz: Summary dashboard | Security: Encrypted data flow Role-based access Audit logging

The Challenge

Interviewing candidates is one of the most critical stages of recruitment. But translating those interviews into structured, useful reports is still a highly manual process. Notes vary from one interviewer to another. Key insights get missed. Comparing candidate responses to job requirements is time-intensive and prone to human bias.
A recruitment tech platform needed a way to automate this process. The goal: quickly generate high-quality interview summaries, highlight gaps between candidate responses and job expectations, and scale the process across high volumes of interviews.

A Day in the Life: Before Our Solution

A recruiter conducts an interview, then spends 20–30 minutes writing a summary. They manually compare what the candidate said to the job description, trying to flag any mismatches or missing details.
If the recruiter is tired or rushed, insights go undocumented. If a hiring manager needs a comparison between multiple candidates, they're often left with subjective or inconsistent notes.

Pain Points:

Interview reports took 20+ minutes per candidate
Manual summaries were inconsistent across recruiters
Gaps between candidate responses and job descriptions were hard to track
Hiring managers lacked clarity in decision support
No scalable way to review interviews at volume

Solution

1. Core Innovation

Tatras Data designed an automated pipeline that transforms raw interview transcripts into structured reports aligned with job requirements.

Transcript Ingestion:: Candidate interviews are transcribed and paired with the job description.
Intelligent Compression: The transcript is shortened while preserving key data points and context.
Prompt Analysis: Different prompts are applied for summary generation, sentiment scoring, and gap detection.
Staged Execution via AWS: The system runs via AWS Step Functions and Lambda to enable failover, parallel processing, and model fallback when needed.
Storage and Access: Results are stored in MongoDB and available to recruiters through an internal dashboard.

2. Key Features

Automated Summarization: Condenses 30-minute interviews into digestible summaries
Job Fit Mapping: Highlights where a candidate aligns — or falls short — of role expectations
Sentiment Analysis: Surfaces tone, confidence, and language trends
Scalable Deployment: 2–3 minute processing via AWS Lambda architecture
Searchable Recordkeeping: Stores results securely for audit and review

3. Workflow Integration

The pipeline now powers post-interview reporting for the recruiting platform. Recruiters receive a structured summary, sentiment score, and job-fit analysis minutes after the interview ends.

Outcomes

✅ 95% accurate summaries in under 3 minutes

📉 >90% reduction in manual report time

💬 Consistent language and tone detection across interviews

🚀 Ready-to-scale framework that supports fallback models and multistage prompts