Tagline
AI-led talent screening platform.
Tech Stack
Python, Django, React, JavaScript, Tensorflow, TailwindCSS, PostgreSQL
Year
2023
SkirmisherAI
A progressive web app for the first pass of hiring. LLM and ML pipelines score each CV against the role, surface interview signals worth probing, and produce a shortlist report the hiring manager can act on without a second pass.
Glimpse of the Project
Hiring Workflow
01 / 0501
Post Vacancy
Employers post roles with structured requirements — title, stack, seniority, must-haves, and a free-form description that the LLM later uses for semantic matching.
Django · PostgreSQL
tap a step to explore →
Case Study
The problem
The first pass of hiring is where good candidates get lost. A single vacancy can pull hundreds of CVs, and screening them by hand is slow, inconsistent, and biased by whoever reads first. Recruiters and hiring managers need a defensible shortlist, not a shuffled inbox.
What I built
A progressive web app where employers post roles with structured requirements and candidates apply with a CV and cover note. An LLM extracts skills, experience and employment gaps from each CV; a scoring model ranks candidates against the role's requirement vector; and the system drafts targeted interview questions plus a narrative suitability report — strengths, risks and fit — that a recruiter can hand straight to the hiring manager.
The hardest decision
Two decisions mattered. First, offline-first: the application flow queues on patchy connections and resyncs, rather than losing a half-finished application — which meant taking on client-side sync and conflict handling instead of a simple online-only form. Second, I paired an LLM for semantic extraction with a separate scoring model rather than letting the LLM rank candidates end-to-end. An LLM-only ranking is a black box you can't defend in a hiring context; keeping scoring explicit keeps the shortlist explainable.
Outcome & what I'd do differently
The suitability report is the part people actually use — it turns a pile of CVs into something shareable. Screening is a domain where calibration and fairness matter more than raw accuracy, so if I extended it I'd add a human-in-the-loop feedback signal and audit scores for bias against employment gaps before trusting the ranking further.
Find Work Screen
Landing Page
Dashboard
Report Generation
Manage Vacancies
Applicant Suitability
Community Page
Contact Page
Terms and Conditions
Designed and Developed by Takinur M