AI-assisted interviews and candidate matching

During my cooperative education placement at SCGC, I worked on AI recruiting tools for UniRecruit. The project connected candidate information, job requirements, interviews, and HR workflows.

The problem

Recruiting involved repeated screening work, disconnected candidate information, and time spent preparing and reviewing interviews. The system needed to help HR assess candidates using the information already available in resumes and job descriptions.

My contribution

I built an AI pre-interview workflow using multiple agents, document retrieval, and context management. I also developed graph-based candidate matching with Neo4j and embeddings, and a live interview assistant with candidate retrieval, session memory, and structured evaluations.

The work extended into backend and cloud services, video processing, job-ad generation, and the application interface. My placement covered May through November 2025.

Retrieval and matching

For the interview workflow, candidate documents and job information became part of the context available to the AI system. The project architecture includes document chunking, embeddings, vector retrieval, session state, and interview processing.

Candidate matching used a different structure. A knowledge graph and embeddings connected skills, experience, and job requirements. The poster describes extraction and normalization, rule-based matching, graph retrieval, and scoring in that workflow.

From an interview to useful records

The platform included virtual pre-interviews, transcription, video analysis, and a live assistant. The poster shows LiveKit for real-time sessions, Redis-backed state and job queues, and cloud services for processing and storage.

I worked on the supporting services and integration so those parts could operate within the recruiting application. The resulting records could support candidate review and later interview stages.

Reported results

The project comparison reports about 70% lower pre-interview cost per session. The poster shows an estimated session cost of 40.29 THB for the HR comparison and 11.77 THB for the AI system. These are the project’s reported comparisons.

The project received 1st Runner-up in the Innovation Category, Oral Presentation, at the KMUTNB Cooperative Education Project Competition in February 2026.

The poster below provides the architecture, interface examples, and cost comparison. The award photo is the selected image of me with the first-runner-up board.