How imchandanmohan/smart-resume-builder works
Project Type & Purpose
• This is a full-stack, AI-driven, voice-enabled resume builder (“Resume Genius”).
• Frontend: Streamlit web app (voice/text input, interactive editing).
• Backend: Python pipeline that ingests job descriptions, runs NLP & ML models (NER, keyword extraction, similarity), scores for ATS, renders a PDF resume, and sends it via email.
Main Architectural Components
a. Frontend Layer
• Streamlit app (Web/app.py + Web/pages/Resume_Builder.py)
• Voice interface module (SpeechRecognition / Whisper)
b. Orchestration & Business Logic
• Pipeline entry points (src/analyse_pipeline.py, src/analyze_jd_focus.py)
• ATS scoring (src/ats_score.py)
• Resume rendering (src/render_resume.py)
• Email sending (email_sender.py or Flask-Mail / SendGrid)
c. Extractor & NLP Services
• Job-description extractor (src/extractors/jd_llm_extractor.py)
• Resume extractor (src/extractors/resume_llm_extractor.py)
• NER service & predictor (src/ner/ner_service.py, src/ner/ner_predictor.py)
• Text cleaning & common utilities (utils/text_cleaner.py, utils/common_functions.py, utils/llm_client.py)
d. Model & Template Artifacts
• Pretrained NER models (artifacts/ner_model/checkpoint-*)
• Resume LaTeX template (artifacts/resume/resume_template.tex.j2)
e. Configuration & Infrastructure
• Config files (config/paths_config.py, config/highlight_config.py, config/KPI_config.py)
• Logging (logs/)
• Dependency management (requirements.txt, setup.sh)
Component Interactions & Data Flow
• User → Frontend: submits job description (text or voice).
• Frontend → Orchestrator: calls pipeline APIs.
• Orchestrator → Extractors: invokes JD extractor & resume extractor (via LLM client).
• Extractors → NER Service: tags entities in resume & JD.
• Pipeline → ATS Scorer: computes ATS-compliance score (TF-IDF, cosine similarity, rule-based).
• Pipeline → Renderer: fills Jinja2 LaTeX template → generates PDF (pdfkit/python-docx).
• Pipeline → Email Sender: posts final PDF to SMTP/SendGrid.
• Throughout: configuration lookup & logging.
Architectural Patterns & Principles
• Layered architecture: UI / Orchestration / NLP services / Persistence (models/templates) / External services.
• Pipeline pattern: discrete stages (extract → tag → score → render → send).
• Single-Responsibility: each module focuses on one task (e.g., ner_service.py only handles NER inference).
• Modular config & utilities: share common functions & client wrappers.