How marcocamilo/ai-resume-alignment-engine works
Project type and purpose
• This is a single‐page, full-stack Python web application (Streamlit) that implements an AI-powered resume alignment engine.
File structure
• Top-level
• app.py: Streamlit entrypoint, glues everything together
• requirements.txt, pyrightconfig.json: Python dependencies and type-checking config
• src/
• modules/
– nlp.py: text preprocessing, tokenization, similarity checks
– utils.py: shared utilities (e.g., file I/O, logging)
• templates/
– prompts.py: prompt templates for evaluation and optimization chains
– meta.py: chain configuration and LangChain instantiation
• description/: sample job-description markdown files
• resume/: sample resume markdown files
• No separate “backend” directory—app.py invokes src modules directly. No persistent database; data flows in memory.
Extract architectural insights from README
• Streamlit UI for user interaction
• Input Processing → AI Analysis (Gemini + LangChain) → Resume Evaluation Chains → Optimization Chains → Visualization (word clouds, tables)
• External service: Google Gemini API
• Core frameworks: Streamlit, LangChain, Google Cloud SDK
Main components and their interactions
a. Components
1) Streamlit Frontend (app.py)
2) Input Handler / Preprocessor (src/modules/nlp.py)
3) Chain Orchestrator (src/templates/meta.py + prompts.py)
4) External LLM Service (Google Gemini API)
5) Postprocessor & Visualizer (word-cloud generation, result formatting)
6) Static Samples (src/description, src/resume)
b. Relationships & data flow
• User ⇨ Streamlit UI ⇨ Input Handler
• Preprocessed text ⇨ LangChain orchestrator
• Orchestrator ⇨ Gemini API calls ⇨ LLM responses
• Responses ⇨ Postprocessor (score calculation, recommendation generation)
• Postprocessor ⇨ Visual components (word clouds, tables) ⇨ Streamlit UI
c. Architectural patterns / design principles
• Single-process service with modular code separation (UI vs. domain logic vs. templates vs. I/O)
• Pipeline style: clear stages (preprocessing → analysis → optimization → visualization)
• Prompt engineering via templated prompts and chain-of-thought orchestration (LangChain)
d. Key technologies / frameworks / libraries
• Streamlit (UI)
• LangChain (chain orchestration)
• Google Gemini API (LLM)
• Python modules: nlp, utils, prompts, meta
• Word-cloud or plotting library for visualization
Detailed breakdown
• Labels for each micro-module (e.g., “Prompt Templates,” “Chain Config,” “Resume Preprocessor”)
• Retry/fallback logic if the Gemini API fails (optional)
• Optional: annotate performance metrics (e.g., “20s average runtime”) on the data flows