The Problem
Preparing for job interviews is a guessing game. Most candidates don't know which skills from their background to emphasise, which gaps a hiring manager will focus on, or how to structure answers to the real questions being asked. Generic interview guides don't help because every role, company, and interviewer is different.
I built this tool to solve my own problem first. then productised it after testing it with a group of friends who were job hunting.
What I Built
CV vs Job Description analysis: the core flow. Paste in a job description, upload or write your CV. GPT-4o performs a gap analysis: which skills and experiences in the JD are present in your CV (and how strongly), which are absent, and which are present but framed in a way that won't read as relevant. The output is a structured match score with specific, actionable rewrites for each weak section.
CV rewriter: one step further. After the gap analysis, users can select any bullet point or section and ask the AI to rewrite it to be more aligned with the target role. All rewrites are grounded in the user's actual experience (the AI can't fabricate credentials). it reframes, quantifies, and sharpens existing content.
Mock interview simulator: once a CV + JD pair is set up, the user can launch a mock interview. GPT-4o plays the interviewer role using a custom prompt that draws from the specific JD, company context (scraped from the company website), and the gaps identified in the analysis. Questions range from standard competency-based questions to role-specific technical questions and the "hard" questions a particular company is known for.
Answer feedback: after each mock answer, the system provides structured feedback: what was strong, what was missing, which interviewer concern the answer should address, and a 1-5 rating. Users can attempt answers multiple times and see improvement.
Preparation plan: a structured 5-day prep plan generated from the analysis, prioritising the highest-impact areas. Each day has specific tasks: questions to practise, topics to review, and connections to draw from past experience.
Technical Highlights
The system is a FastAPI backend with Next.js frontend. CV parsing handles PDF and DOCX formats (pdfminer + python-docx). JD processing normalises the free-text job description into structured skill and requirement categories using a classification chain before the gap analysis.
Mock interviews use a streamed completion. interview questions and feedback appear token by token, making the experience feel more like a real conversation. The interview state (question history, answers, feedback) is stored in PostgreSQL so sessions can be resumed.
I built the ATS scan feature later. it checks the CV for formatting issues that applicant tracking systems commonly reject (tables, graphics, unusual section headers, missing keywords in standard positions). This was one of the most-requested features based on user feedback.
Outcome
Launched publicly after 3 months of development. 420 active users within the first 6 weeks, primarily through organic LinkedIn posts and word of mouth among job seekers.
In a survey of 80 users who tracked their outcomes: those who completed a full prep cycle (CV rewrite + 5+ mock interviews) reported a 3.1× improvement in their interview-to-offer ratio compared to their previous job search cycle. Interesting caveat: users who only did the CV analysis without mock practice showed much smaller improvement.
Revenue: $1,200 MRR at 3 months on a freemium model (free CV analysis, paid mock interviews). Still running and growing.