How kvj-harsha/smarthire works
Project Type and Purpose
• This is a full-stack AI-powered data pipeline application with two primary facets:
• A backend “orchestrator” pipeline that ingests candidate and job data, invokes AI modules (agents), and produces Talent Intelligence Reports (TIRs).
• A Streamlit-based frontend dashboard for visualizing profiles, match analyses, behavioral insights, and market intelligence.
File Structure
• Top-level entries:
• agents/ – four AI modules (candidate_profiler, assessment_designer, behavioral_analyzer, market_optimizer)
• app/orchestrator.py – main pipeline runner
• app.py and pages/ – Streamlit dashboard entry point and multipage views (assessment, behaviour, data-viewer, market, profile)
• data/ – raw input JSONs (resume, LinkedIn, GitHub, LeetCode, JD, textual evidence)
• talent-intelligence-report/ – generated JSON outputs (orchestrated data and final TIRs)
• requirements.txt and .env – dependencies and environment variables (e.g., GROQ_API_KEY)
Extract Insights from README
• Backend flow: load data → run profiling, behavioral analysis, assessment design, market intelligence → save both “_orchestrated” and final “TIR_” outputs.
• Frontend: streamlit run app.py uses pages/ to display results.
• Technologies: Python, Groq LLM API, python-dotenv, Streamlit.
Main Components
a. Data Sources Layer
• Local JSON data files (data/, representing resume, LinkedIn, GitHub, LeetCode, JD, candidate_text)
• Environment variables store API keys
b. Orchestrator Pipeline
• app/orchestrator.py
• Invokes agent modules sequentially or in parallel:
• Candidate Profiler (agents/candidate_profiler.py)
• Behavioral Analyzer (agents/behavioral_analyzer.py)
• Assessment Designer (agents/assessment_designer.py)
• Market Optimizer (agents/market_optimizer.py)
• Calls external Groq LLM API for AI processing
• Outputs saved to talent-intelligence-report/
c. Storage Layer
• talent-intelligence-report/ (orchestrated and TIR outputs)
d. Visualization Layer (Streamlit Dashboard)
• app.py (Streamlit entry)
• pages/… (five pages for each functional area)
• Reads from talent-intelligence-report/ or data/ for live rendering
e. External Service
• Groq LLM API (AI inference engine)
Relationships and Data Flow
• User interacts with Dashboard, which triggers page logic to fetch stored outputs
Architectural Patterns and Principles
• Pipeline orchestration pattern: sequential/parallel execution of discrete “agent” components
• Clear separation of concerns: data ingestion, AI modules, reporting storage, visualization
• Config via environment (.env) and requirements.txt for dependency management
• Modular code layout to enable easy extension (add new agents or pages)
Technologies and Frameworks to Label
• Python 3.x, python-dotenv
• Streamlit (frontend)
• Local filesystem for JSON storage
Specific Elements
• Clear labels for each module (with file paths or names)
• Blue for backend components
• Green for data stores
• Orange for external services
• Purple for frontend/UI