About — IntelliHire

Why This Stack?

One API key. One repository. One deployment platform. Maximum intelligence.

🧠

AI Does Everything

Resume parsing, adaptive MCQ generation, code review, HR simulation, behavioral scoring — all powered by a single Anthropic API key via direct fetch() calls from the browser. No middleware needed.

📦

GitHub as Backend

The GitHub repository is both the codebase and the CI/CD pipeline. GitHub Actions auto-builds and deploys every push. GitHub Pages serves the static frontend globally with HTTPS out of the box.

⚡

Zero Infrastructure

No FastAPI, no Docker, no cloud VMs, no databases. All state lives in localStorage and sessionStorage. All intelligence lives in AI. The entire platform is a folder of HTML, CSS, and JS files.

🔐

API Key Security

The ANTHROPIC_API_KEY is stored as a GitHub Actions secret and injected at build time into a config file — never committed to the repo or exposed in public source. Requests go directly to api.anthropic.com.

Core Technology

🤖

AI API

Anthropic · All AI logic

📄

Vanilla JS

ES2023 · No frameworks

🎨

HTML + CSS

Static frontend

🐙

GitHub Pages

Hosting + CDN

⚙️

GitHub Actions

CI/CD pipeline

💾

localStorage

Client-side state

Global Stack

AI / LLM

Anthropic AI GPT-4oclaude-haiku-4-5 (fast tasks)Streaming API

Frontend

HTML5CSS3 (CSS Variables)Vanilla JS (ES2023)Fetch APIWeb Speech APIMediaDevices (webcam)

State

localStorage (session)sessionStorage (temp)No external DB

Hosting

GitHub PagesHTTPS (auto)Global CDN

CI/CD

GitHub ActionsSecret injection (ANTHROPIC_API_KEY)Auto-deploy on push to main

Libraries

PDF.js (PDF parsing)Monaco Editor (code)Chart.js (visualizations)Marked.js (markdown)

Platform Modules

Five AI-powered hiring stages, all driven by AI

MODULE 01

ATS Resume Analyzer

AI reads your resume + JD, scores it, rewrites weak bullets, generates a polished PDF.

● Live

MODULE 02

Aptitude Round

AI generates adaptive MCQs in real time, adjusting difficulty based on your answers.

● Live

MODULE 03

Technical Round

AI acts as a senior engineer — reviews your code, asks follow-ups, scores logic and style.

◐ In Progress

MODULE 04

HR Simulator

AI simulates an HR interviewer using STAR framework, scores communication and clarity.

◐ Upcoming

MODULE 05

Behavioral Analysis

AI analyzes webcam-captured answers for tone, filler words, pacing, and confidence signals.

◐ Upcoming

Module Deep Dives

How each module uses the AI API

01 — Resume Live

02 — Aptitude Live

03 — Technical WIP

04 — HR Soon

05 — Behavioral Soon

AI Flow — Module 1

Upload Resume (PDF.js)

→

AI: Parse + Score

→

AI: Rewrite Bullets

→

AI: Suggestions

→

Generate PDF (JS)

📝

AI Resume Parser

Anthropic API — Extraction

A structured system prompt instructs AI to extract name, email, phone, GitHub, education, experience, skills, and projects from raw PDF text. Returns a clean JSON object — no regex required.

Model

GPT-4oJSON mode output

🔍

AI ATS Scorer

Anthropic API — Scoring

AI receives the resume JSON + job description text. It returns a 0–100 ATS score with six sub-scores: keyword match, skills match, experience relevance, formatting, readability, section completeness — all evaluated by language understanding.

Model

GPT-4oStructured JSON response

✍️

Bullet Rewriter

Anthropic API — Enhancement

Each weak experience bullet is sent to AI with a STAR-method prompt. AI rewrites it to start with an action verb, adds metrics, and keeps it under 25 words. Temperature 0.7 for variety.

Model

claude-haiku-4-5Temp 0.7

💡

Improvement Suggestions

Anthropic API — Advisory

AI receives the gap between resume skills and JD requirements. It returns 8 prioritized improvement suggestions — missing keywords to add, sections to strengthen, formatting fixes. Temperature 0.3 for factual advice.

Model

GPT-4oTemp 0.3

Prompt Engineering — Module 1

System Prompt — Resume Parser

Role: Expert resume parser. Extract structured data. Return only valid JSON.

Schema: {name, email, phone, github, education[], experience[], skills[], projects[]}

Rules: No markdown, no explanation. If field missing, return null. Dates as ISO strings.

System Prompt — Bullet Rewriter

Role: Senior resume coach. Rewrite bullets using STAR method. Start with action verb.

Rules: Add quantified metrics (%, $, scale). Max 25 words. Return JSON array of strings.

Input: "Worked on backend APIs"

Output: "Architected 5 REST endpoints reducing client latency by 35%, serving 50k daily active users"

Streaming Pattern (JS fetch)

const res = await fetch('https://api.anthropic.com/v1/messages', {

headers: { 'x-api-key': ANTHROPIC_API_KEY, 'anthropic-version': '2023-06-01' },

body: JSON.stringify({ model: 'GPT-4o', stream: true, ... })

});

API / Library	Provider	Used For	Cost
GPT-4o	Anthropic	Resume parsing, ATS scoring, improvement suggestions	Pay-per-token
claude-haiku-4-5	Anthropic	Bullet rewrites (faster, cheaper for high-volume)	Pay-per-token
PDF.js	Mozilla	Extract text from uploaded PDF in browser	Free (open source)
jsPDF	JS lib	Generate ATS-safe resume PDF client-side	Free (open source)
GitHub Pages	GitHub	Host and serve the entire frontend	Free
GitHub Actions	GitHub	CI/CD — build, inject API key secret, deploy	Free (public repo)

LLM

Anthropic GPT-4oclaude-haiku-4-5

Frontend

HTML5CSS3Vanilla JS (ES2023)Fetch API (streaming)

Parsing

PDF.js (browser PDF text)AI JSON mode (NER)

PDF Gen

jsPDF (ATS-safe, client-side)

State

localStorage (resume + analysis cache)

Deploy

GitHub PagesGitHub Actions (secret: ANTHROPIC_API_KEY)

AI Flow — Module 2

Select Domain + Level

→

AI: Generate MCQ

→

User Answers

→

AI: Evaluate + Next Q

→

Score Report

🎯

Adaptive Question Generator

Anthropic API — Dynamic MCQ

AI generates each question fresh — never from a static bank. The prompt includes: topic domain, target difficulty (1–10), candidate's answer history, and a rule against repeating question types. Produces one MCQ with 4 options and the correct answer tagged.

claude-haiku-4-5Low latencyJSON output

📊

Difficulty Adjuster

Anthropic API — Adaptive Logic

After each answer, AI evaluates correctness and adjusts difficulty. Correct answer → increase difficulty by 1. Wrong answer → decrease by 1. AI also provides a short explanation of the correct answer for learning mode.

claude-haiku-4-5Adaptive IRT-style

🔊

Voice Mode

Web Speech API + AI

Web Speech API reads questions aloud (TTS) and transcribes spoken answers (STT). Transcribed text is sent to AI for evaluation. Enables hands-free aptitude testing that simulates real verbal screening.

Web Speech APIAI evaluation

Prompt Engineering — Module 2

MCQ Generation Prompt

System: Expert aptitude test designer. Generate exactly ONE multiple-choice question.

Input: {domain: "Data Structures", difficulty: 6, previousTopics: ["arrays","sorting"]}

Output: {question, options: [A,B,C,D], correct: "B", topic, difficulty, explanation}

Rules: No repeated topics from previousTopics. Only valid JSON. No markdown fences.

API / Library	Provider	Used For	Cost
claude-haiku-4-5	Anthropic	Fast MCQ generation + answer evaluation	Pay-per-token
Web Speech API (TTS)	Browser	Read questions aloud in voice mode	Free (native browser)
Web Speech API (STT)	Browser	Transcribe spoken answers for evaluation	Free (native browser)
Chart.js	JS lib	Topic-wise performance radar chart in report	Free
GitHub Pages	GitHub	Hosting	Free

LLM

claude-haiku-4-5 (speed-optimized)

Voice

Web Speech API — TTS (question readout)Web Speech API — STT (answer capture)

Frontend

Vanilla JSCSS3 animationsChart.js (radar)

State

sessionStorage (active quiz session)localStorage (score history)

Deploy

GitHub PagesGitHub Actions CI

AI Flow — Module 3

Select Language + Domain

→

AI: Pose Problem

→

Write Code (Monaco)

→

AI: Review Code

→

AI: Follow-up Q

→

Technical Score

💻

Problem Poser

Anthropic API — Interviewer

AI acts as a senior engineer and poses a domain-appropriate coding problem. It calibrates difficulty based on candidate's resume skills (parsed in Module 1). Problem includes constraints, examples, and expected time complexity.

GPT-4oResume-aware

🔎

Code Reviewer

Anthropic API — Evaluation

AI reviews submitted code for correctness, time/space complexity, edge case handling, naming conventions, and readability. Returns a structured score (0–100) with line-by-line feedback rendered inline in Monaco Editor.

GPT-4oStructured review

🗣️

Follow-up Interviewer

Anthropic API — Dialogue

After code review, AI asks 2–3 natural follow-up questions: "Why did you choose a hashmap here?", "What if the input is sorted?", "How would you scale this to 10M records?" Simulates a real technical conversation.

GPT-4oConversational

API / Library	Provider	Used For	Cost
GPT-4o	Anthropic	Problem generation, code review, follow-up dialogue	Pay-per-token
Monaco Editor	Microsoft	In-browser code editor (VS Code engine, CDN)	Free (CDN)
GitHub Pages	GitHub	Hosting	Free

LLM

GPT-4o (problem + review)

Editor

Monaco Editor (via CDN)Syntax highlightingInline decorations (feedback)

Frontend

Vanilla JSMarked.js (render AI markdown feedback)

State

sessionStorage (active problem + conversation history)

Deploy

GitHub PagesGitHub Actions CI

AI Flow — Module 4

Resume Context Loaded

→

AI: HR Persona

→

Candidate Responds

→

AI: STAR Scoring

→

Communication Report

🧑‍💼

HR Persona Engine

Anthropic API — Role Play

AI is given a system prompt that defines it as a friendly but rigorous HR interviewer at a specific company type (startup, MNC, product company). It asks STAR-format behavioral questions tailored to the candidate's resume and target role.

GPT-4oPersona prompt

⭐

STAR Evaluator

Anthropic API — Scoring

After each answer, AI scores STAR completeness: Situation (did they set context?), Task (what was their role?), Action (specific steps taken?), Result (quantified outcome?). Returns a JSON score per component plus improvement tip.

GPT-4o4-axis STAR score

💬

Filler Word Detector

JS + AI NLP

Web Speech API transcribes spoken answers. A JS regex pass counts filler words (um, uh, like, you know, basically). AI then evaluates fluency, clarity, and confidence in the transcript. Combined score weighted 30% fillers, 70% content.

Web Speech APIAI fluency

API / Library	Provider	Used For	Cost
GPT-4o	Anthropic	HR persona, question generation, STAR scoring, fluency eval	Pay-per-token
Web Speech API	Browser	Transcribe spoken HR answers	Free
GitHub Pages	GitHub	Hosting	Free

LLM

GPT-4o (HR persona + STAR scorer)

Voice

Web Speech API TTS (HR questions)Web Speech API STT (candidate answers)

NLP

JS regex (filler word count)AI (fluency + content scoring)

State

sessionStorage (interview transcript + turns)

Deploy

GitHub PagesGitHub Actions CI

AI Flow — Module 5

Webcam + Mic Active

→

Record Answer (30s)

→

AI: Transcript Analysis

→

AI: Tone + Confidence

→

Behavioral Score

🎥

Video Answer Recorder

MediaDevices API — Capture

Native browser MediaDevices API captures webcam + microphone. No external service needed. Video is stored in memory as a Blob. Web Speech API simultaneously transcribes the spoken answer in real time.

MediaDevices.getUserMedia()MediaRecorder API

📊

Behavioral Analyzer

Anthropic API — Multi-dim Analysis

AI receives the full speech transcript and evaluates: communication clarity (30%), confidence signals (25%), structured thinking (25%), vocabulary range (20%). Returns scores per axis and a natural-language coaching summary.

GPT-4o4-axis analysis

🗣️

Pacing + Tone Detector

JS + AI NLP

JS calculates speech rate (words per minute) from transcript timestamps. Filler word frequency is counted via regex. AI evaluates tone (assertive/hesitant/enthusiastic) and overall delivery quality from the full transcript text.

WPM calc (JS)Tone (AI)

API / Library	Provider	Used For	Cost
GPT-4o	Anthropic	Transcript behavioral analysis, tone + confidence scoring	Pay-per-token
MediaDevices API	Browser	Webcam + microphone capture	Free (native)
MediaRecorder API	Browser	Record video answer as Blob in memory	Free (native)
Web Speech API	Browser	Real-time transcription of spoken answer	Free (native)
Chart.js	JS lib	Behavioral score radar chart in report	Free
GitHub Pages	GitHub	Hosting	Free

LLM

GPT-4o (transcript + behavioral scorer)

Video

MediaDevices.getUserMedia()MediaRecorder (Blob in memory)

Voice

Web Speech API STT (real-time transcript)

NLP (JS)

WPM calculatorFiller word regexVocabulary range counter

Charts

Chart.js radar (behavioral dimensions)

Deploy

GitHub PagesGitHub Actions CI

Deployment Architecture

How the entire platform ships from a single GitHub repo

Source

GitHub Repository (public)All HTML/CSS/JS committedNo server code

Secret

ANTHROPIC_API_KEY stored as GitHub Actions SecretInjected at build time into config.jsNever committed to repo

CI/CD

GitHub Actions workflow (.github/workflows/deploy.yml)Triggers on push to mainBuilds config.js with secretDeploys to gh-pages branch

Hosting

GitHub Pages (static)HTTPS enforced (auto)Global CDNCustom domain support

Runtime

Browser makes direct fetch() to api.anthropic.comNo proxy, no backend, no serverCORS allowed by Anthropic API

GitHub Actions Deploy Workflow (simplified)

on: push to main

steps:

1. echo "const ANTHROPIC_API_KEY = '${{ secrets.ANTHROPIC_API_KEY }}';" > config.js

2. Deploy all files to gh-pages branch via actions/deploy-pages

3. Site live at https://student-cybrarians.github.io/intellihire/

Project Roadmap

Development milestones for all 5 modules

Phase 1 · Completed

Module 01 — ATS Resume Analyzer

AI-powered resume parsing, ATS scoring, bullet rewriting, jsPDF resume generation. Live on GitHub Pages.

Phase 2 · In Progress

Module 02 — Aptitude Round

AI adaptive MCQ generation, difficulty adjustment, Web Speech API voice mode, topic-wise Chart.js report.

Phase 3 · Upcoming

Module 03 — Technical Round

AI code reviewer, Monaco Editor integration, follow-up dialogue, language-agnostic problem generation.

Phase 4 · Upcoming

Module 04 — HR Simulator

AI HR persona, STAR method scoring, filler word detection via Web Speech API + JS regex, fluency report.

Phase 5 · Upcoming

Module 05 — Behavioral Analysis + Final Report

MediaDevices webcam capture, real-time STT transcript, AI behavioral scoring, 4-axis radar report. Full platform integration.

Built By

Student-Cybrarians

B.Tech CSE (AI/ML) — Major Project

⭐ GitHub Repository 🚀 Live Site

How IntelliHireActually Works

AI Does Everything

GitHub as Backend

Zero Infrastructure

API Key Security

How IntelliHire
Actually Works