Intelligent Job Matching for IT Recruitment

Table of Contents Toggle OVERVIEWBUSSINESS CHALLENGESCOPE OF WORKOUTCOMESSystem PerformanceEfficiency Improvements OVERVIEW Product: Interview Analysis System – automated transcription & evaluation reports Domain: HR & technical recruitment Scope: Audio chunking, Whisper transcription, GPT-4 evaluation & reports AX expertise: Speech-to-Text · Gen-AI · Audio Processing · NLP Engagement: End-to-end product build (audio pipeline + LLM analysis) BUSSINESS CHALLENGE A technical-recruitment firm ran 150+ interviews monthly with inconsistent scoring across interviewers. They wanted automated transcription, structured evaluation and standard feedback per session. SCOPE OF WORK Audio chunking and pre-processing pipeline (FFmpeg) Whisper-based transcription with technical-term tuning Speaker separation for Q&A turn-by-turn analysis GPT-4 content evaluation + technical competency scoring Structured report generation with key-moment highlights OUTCOMES 87.5% less analysis time From 1–2 hours to ~15 minutes per interview. 95% scoring consistency Inter-rater agreement up from a 65% baseline. 2× interviewer output From 3 to 6 productive interviews per day. 14-day hiring cycle Whole funnel cut from 21 days end-to-end. System Performance Transcription Accuracy: 98% for clear audio Analysis Generation: 5-7 minutes per interview System Uptime: 99.9% Efficiency Improvements

Table of Contents Toggle OVERVIEWBUSSINESS CHALLENGESCOPE OF WORKOUTCOMES OVERVIEW Product: Intelligent Job Matching – semantic CV-to-JD matching at recruitment scale Domain: HR & IT recruitment Scope: CV/JD parsing, OpenAI embedding pipeline, Pinecone vector search AX expertise: Embeddings · Semantic Search · NLP · Custom Scoring Engagement: ML build + AWS deployment (FastAPI + EKS) BUSSINESS CHALLENGE An IT recruitment agency processed 5,000–7,000 technical CVs monthly. Keyword matching missed strong candidates and over-prioritised buzzwords. They needed semantic, skills-aware matching at scale. SCOPE OF WORK Industrial camera and synchronised lighting hardware setup Custom OCR model tuned to steel-plate engraving conditions Real-time image-quality verification with re-capture loop Schema-matched extraction and validation before write Automated ingestion into existing Oracle database Exception alerting and monitoring dashboards OUTCOMES Hands-free capture Plates logged automatically as they pass inspection. Schema-validated writes Extracted text matched against expected formats. Closed-loop monitoring One operator supervises the model for a month, not every shift. Direct Oracle integration No middleware – the database is the system of record.  

Table of Contents Toggle OVERVIEWBUSSINESS CHALLENGESCOPE OF WORKOUTCOMES OVERVIEW Product: Steel Plate Image-to-Text – automated OCR & Oracle ingestion on factory floor Domain: Steel manufacturing Scope: Industrial camera capture, custom OCR model, Oracle DB integration AX expertise: Computer Vision · OCR · Hardware Integration Engagement: End-to-end vision system (hardware + ML + database) BUSSINESS CHALLENGE A steel-processing facility logged plate IDs into Oracle manually – slow, error-prone and a bottleneck on throughput. They wanted automated capture with industrial cameras and direct database integration. SCOPE OF WORK Industrial camera and synchronised lighting hardware setup Custom OCR model tuned to steel-plate engraving conditions Real-time image-quality verification with re-capture loop Schema-matched extraction and validation before write Automated ingestion into existing Oracle database Exception alerting and monitoring dashboards OUTCOMES Hands-free capture Plates logged automatically as they pass inspection. Schema-validated writes Extracted text matched against expected formats. Closed-loop monitoring One operator supervises the model for a month, not every shift. Direct Oracle integration No middleware – the database is the system of record.