I built a tool that tells you exactly why you'll be rejected — before you apply
I have watched classmates spend entire weekends tailoring resumes for roles they were never going to get past the first filter.
The rejection comes two weeks later. No reason. No feedback. Just silence.
That silence is the problem I decided to fix.
The Tool Everyone Builds vs The Tool Everyone Needs
Everyone builds a resume scorer.
Upload resume → get 72% match → feel bad → close the tab.
That number tells you nothing. A hiring manager does not think in percentages. They think:
"This person has never touched Kubernetes. We need someone who can run distributed training on SageMaker on day one. Next."
That thought happens in 30 seconds. Your application is gone.
Rejected Before Applying is the tool that says that thought out loud. Before you click submit. While you still have time to do something about it.
How It Actually Works
This is not a single LLM call with a smart prompt. It is a two-layer pipeline where the AI earns its role.
Layer 1 — Deterministic Keyword Engine
Before the AI touches anything, a rule-based engine scans both your resume and the job description against a curated vocabulary of 60 technical skills.
No hallucination. No guessing. Hard evidence.
This evidence is passed directly into the AI prompt. The model does not decide what is missing. The engine already knows. The model explains it.
Layer 2 — AI Hiring Manager (Llama-3.1-8B)
Grounded in the keyword evidence above, the model produces five things:
1. The rejection diagnosis Not "you lack infrastructure skills." This:
"You have zero production deployment experience. Every project on your resume is research. This role requires someone who can run distributed training on SageMaker on day one. You are not that person yet."
2. Top 3 skill gaps — by exact technology name Not "cloud skills." Kubernetes. Not "data tools." Apache Kafka.
3. Specific projects to build Not "get more experience." "Deploy a fine-tuned BERT model on Kubernetes with an MLflow tracking server. Document it on GitHub."
4. A 30-day plan — week by week Week 1 through Week 4. Concrete. Actionable. No filler.
5. Fix My Resume — 3 bullets to add after building those projects
"Deployed NLP model serving 500 req/s on Kubernetes cluster using Docker and MLflow, reducing inference latency by 40%."
The app does not just identify the problem. It gives you the exact path out of it.
Why a Small Model Is the Right Call
The hard reasoning — what skills are missing — is done deterministically. Zero hallucination risk on the most important part of the output.
The model handles what models are actually good at: explanation, empathy, planning, and writing resume bullets that sound human.
An 8B model does all of that well.
This is what Build Small means to me: use the smallest model that earns its place in the pipeline. Not the biggest model you can find.
Upload Your Resume. Find Out The Truth.
PDF. DOCX. TXT. Paste or upload.
The report takes 20 seconds.
The truth might sting. But it is infinitely more useful than silence.
👉 Try Rejected Before Applying
Built for the 🤗 Build Small Hackathon.
Small model. Engineered pipeline. A problem every student in this hackathon has personally felt.