How InterviewProof Works
InterviewProof uses an evidence-based diagnostic pipeline to identify the specific gaps between your profile and a role's requirements. Every score is calculated by code, not generated by AI.
The Pipeline
Your resume and job description pass through five stages, each building on the last.
Extract
An LLM parses your resume and job description into typed schemas -- skills taxonomy, seniority signals, requirement classification, and experience mapping. This is semantic extraction, not keyword matching.
The extraction model understands context: "led a team of 8" is a leadership signal, not just a number. Every field is structured for downstream analysis.
Retrieve
Your extracted profile is converted into 1536-dimensional embeddings and matched against a curated knowledge base of 50+ rubrics using pgvector cosine similarity search. This is Retrieval-Augmented Generation (RAG) -- grounding AI output in domain-specific data.
Retrieval is contextual: the system selects rubric chunks based on role type, round format, and company tier -- not random sampling.
Analyze
The LLM performs a gap analysis grounded in the retrieved rubric context (RAG), not just its general training knowledge. It compares your profile against real hiring criteria to identify mismatches, missing evidence, and risk areas.
Output is enforced via strict JSON schema -- structured category scores, ranked risks, coaching insights, and recruiter signals. No freeform text that can drift or hallucinate.
Score
A deterministic scoring engine calculates your readiness score using weighted dimensions. The AI does not generate scores -- code does. This separation is an architectural decision, not an accident.
Reproducible, auditable, and fair. Same inputs always produce the same score. The scoring formula is versioned and stored alongside every result.
Diagnose
Generate a comprehensive diagnostic: risk rankings, interview questions calibrated to your gaps, a study plan, coaching tips, and recruiter-perspective insights.
Every recommendation traces back to specific retrieved rubric chunks and evidence gaps -- not hallucinated advice. You can see the data trail behind each insight.
Retrieval-Augmented Generation
Most AI tools pass your resume straight to a language model and hope for the best. InterviewProof uses a RAG architecture -- grounding every analysis in domain-specific data, not generic LLM knowledge.
How RAG works
Before the AI analyzes your fit, it retrieves the most relevant hiring criteria from a curated knowledge base. This means the analysis is anchored to real rubrics and evaluation frameworks -- not improvised from the model's training data.
Resume + JD
Your inputs
Embeddings
1536-dim vectors
Vector Search
pgvector cosine similarity
Retrieved Rubrics
Domain-specific context
Grounded Analysis
RAG-powered output
Why this matters for you
Generic AI can tell you to "quantify your achievements." RAG-grounded AI can tell you that for a Senior Backend Engineer role at a Series B startup, the hiring rubric specifically weights system design ownership and on-call experience -- and your resume is missing both. The difference is actionable specificity.
50+
Curated Rubrics
Hand-built, not scraped
1536
Embedding Dimensions
High-fidelity semantic matching
0
Hallucinated Insights
Every claim is evidence-backed
6 Scoring Dimensions
Your readiness score is a weighted composite of five measurable dimensions, plus a company difficulty adjustment.
Hard Match
35%How well your technical skills, tools, and qualifications align with the explicit requirements in the job description.
Evidence Depth
25%Whether your resume provides concrete, quantified proof of your claims -- metrics, outcomes, and specific accomplishments rather than vague assertions.
Round Readiness
20%Your preparation level for the specific interview round type -- technical, behavioral, or research/ML formats.
Clarity
10%How effectively your resume communicates your value -- structure, conciseness, and whether a recruiter can quickly parse your fit.
Company Proxy
10%Company-specific factors including tier difficulty, culture signals, and how your prior experience maps to their expectations.
Company Difficulty Adjustment
A 1.0x to 1.5x multiplier applied based on company tier (FAANG+, Big Tech, Mid-Market, etc.) that adjusts scoring thresholds to reflect the higher bar at more competitive companies.
The Deterministic Principle
Many AI tools generate scores by asking a language model to "rate this candidate from 1-100." That approach is inherently unreliable -- the same input can produce different scores on different runs, and there is no way to audit why a particular number was chosen.
InterviewProof takes a fundamentally different approach: the AI analyzes, the code scores.
The language model serves as the analyst -- it extracts structured data, retrieves relevant rubrics, and produces a detailed gap analysis with category-level assessments. But the final readiness score is calculated by a deterministic scoring engine: a set of weighted formulas implemented in code.
This ensures three critical properties:
Reproducibility
Same inputs always produce the same score. No variance between runs.
Auditability
Every score breakdown is stored and can be inspected. You can see exactly which dimensions contributed what.
Fairness
No LLM mood swings or prompt sensitivity. The scoring formula treats every candidate consistently.
Data Sources
InterviewProof's analysis is grounded in real interview data, not generic advice. Every data source below is embedded as 1536-dimensional vectors in a pgvector database, enabling contextual retrieval based on your specific role type, round format, and company tier.
50+ Hiring Rubrics
Hand-curated evaluation criteria used by companies across tiers, from startups to FAANG+, covering technical, behavioral, and research/ML interview formats. Each rubric is chunked and embedded for precise vector retrieval.
Recruiter Playbooks
Real screening criteria and evaluation frameworks that recruiters and hiring managers use to filter candidates at each stage. Embedded alongside rubrics so the system retrieves the right playbook for your specific interview context.
Interview Question Banks
Role-specific question archetypes mapped to competency areas, ensuring practice questions target your actual weak points -- not generic interview prep.
Company Tier Intelligence
Difficulty calibration data that adjusts expectations based on the company's hiring bar, interview style, and competitive landscape. Retrieval adapts dynamically to company tier.
See it in action
Upload your resume and a job description to get a full diagnostic. You'll see every score, every risk, and every recommendation -- with full transparency into how each was derived.