SkillHunt

Association

SkillHunt is associated with Tecciance.

What it is

SkillHunt is an AI-powered interview platform hosted at skillhunt.net. The core product uses LLMs to evaluate interview-related content by injecting dynamic candidate, assessment, and role data into pre-built prompt templates.

The important engineering problem was making that AI layer behave like a product system, not a demo: consistent prompts, predictable grading, scalable bulk processing, reliable deployments, and reporting flows that hiring teams could actually use.

Surface-level view on what I worked on

flowchart LR
  Input[Interview and candidate data] --> Prompts[Dynamic prompt templates]
  Prompts --> LLM[LLM evaluation]
  LLM --> Grader[Core grading algorithm]
  Grader --> Reports[Reports and emails]
  Bulk[Bulk processing queue] --> Workers[Concurrent async jobs]
  Workers --> Grader
  CI[CI/CD pipeline] --> AWS[AWS EC2 and serverless]

What I built

I worked across the full stack: frontend reporting flows in Next.js, backend APIs in Spring Boot, PostgreSQL-backed persistence, CI/CD pipelines, and AWS deployment work across EC2 and serverless services.

On the AI side, I optimized the prompt workflow and helped build the core grading algorithm that integrated with the LLM responses. The grading layer had to turn generated model output into structured, repeatable evaluation data that could be shown in reports and downstream communication.

I also implemented bulk processing with asynchronous concurrent execution, so larger batches of interview data could be processed without blocking user-facing flows or requiring manual step-by-step handling.

Product surfaces

The frontend work focused on making the AI output useful after generation. I built and fixed report-generation views, handled dynamic report data, and worked on email templates that could adapt to each candidate or assessment state.

That meant connecting backend evaluation results to clean frontend presentation and communication flows, rather than leaving the AI output as raw system data.

Infrastructure and delivery

I worked on CI/CD and deployment setup for AWS EC2 servers, and also contributed to serverless architecture using AWS Lambda, S3, and API Gateway. The goal was to keep deployments repeatable while moving heavier or bursty work into infrastructure that could scale without permanently running extra capacity.

What it changed

Prompt optimization improved model accuracy by about 15% and reduced response latency by about 20%. The bulk processing work improved throughput by roughly 40% and reduced manual processing time from minutes to seconds.

The AWS serverless module reduced infrastructure cost by about 25%, while the EC2 deployment setup helped maintain 99% uptime across development and production environments.