Full-stack AI MVP that matches a CV to a job description and generates an ATS-friendly, role-aligned resume using structured LLM reasoning — with explainable and hallucination-aware improvements.
Run locally • Screenshots • Architecture • AI Workflow
The system analyzes a candidate’s CV against a job description and generates a personalized, role-aligned resume using structured LLM reasoning.
Designed for job seekers who want data-driven resume personalization instead of manual rewriting, helping produce clearer, role-aligned, and ATS-friendly resumes.
- FastAPI backend with modular API architecture
- Structured CV ↔ Job matching pipeline
- LLM-based resume and job description analysis
- Explainable AI enhancement (no fabricated experience)
- Skill match and gap analysis
- PDF & DOCX export support
- Streamlit prototype UI used for rapid validation before migration to a production Next.js frontend
This project represents a functional MVP built to validate an AI-driven resume personalization workflow.
Focus of this version:
- validating structured LLM reasoning pipelines
- testing real user interaction flows
- proving feasibility before production-scale architecture
This repository represents my personal implementation and extension of a collaborative AI training project.
Primary contributions:
- Building the AI orchestration workflow
- Prompt engineering and enhancement logic
- Structuring data schemas using Pydantic models
- Implementing frontend interaction flows for AI evaluation
- Preparing the system architecture for real-world integration
Many junior and mid-level candidates struggle to tailor their resumes for specific roles.
They often rely on guesswork and spend significant time rewriting CVs without understanding recruiter or ATS expectations.
Upload CV + Job Description
→ Structured Analysis
→ AI Alignment & Enhancement
→ Transparent Improvement Report
→ Export optimized resume
- Ensuring factual resume enhancement without hallucinated experience
- Converting unstructured CV text into reliable structured schemas
- Maintaining alignment between job requirements and generated output
- Designing explainable AI outputs instead of black-box rewriting
Pipeline
Structure → Reason → Rewrite → Review → Render
flowchart TB
U[User] --> UI[Streamlit UI]
U --> JD[Job Description Input]
UI --> API1[Parse Endpoint]
UI --> API2[Enhance Endpoint]
UI --> API3[Export Endpoint]
API1 --> API[FastAPI Backend]
API2 --> API
API3 --> API
API --> P[Parsing Layer]
P --> S[Schema Models]
S --> G[Orchestration Graph]
G --> M[Mapping Node]
M --> E[Enhancement Nodes]
E --> F[Formatting Node]
F --> R[Report Node]
G --> LLM[LLM Provider OpenRouter]
LLM --> G
R --> API
API --> UI
UI --> O[Preview and Export Output]
- Resume & job parsing endpoints
- Enhancement and mapping workflows
- Export services (PDF/DOCX)
- LangChain / LangGraph orchestration
Streamlit-based interface used for workflow validation and user interaction testing.
The frontend layer was designed and implemented as part of the MVP to simulate real user interaction with the AI system.
Key aspects:
- User-driven CV and job description workflow
- Real-time enhancement feedback
- Structured result visualization
- UX focused on clarity and explainability of AI changes
Streamlit was intentionally used for rapid product validation before migrating to a production-grade Next.js interface.
Pydantic schemas define a unified contract for:
- Resume
- Experience
- Skills
- Education
- Certifications
- JobDescription
- MappingResult
Upload → Text Extraction → Structured Schemas
Mapping → Score Evaluation → Enhancement → Formatting → Report Generation
- Python
- FastAPI
- Streamlit
- LangChain / LangGraph
- Pydantic
- LLM APIs (OpenAI / Gemini)
- REST API Architecture
| Mode | Description |
|---|---|
| legacy | Full enhancement in a single response |
| incremental | Streamed enhancement with real-time updates |
- Clone Repository
git clone <repo-url>
cd ai-profile-personalization-engine- Backend Setup python -m venv .venv .venv\Scripts\activate pip install -r requirements.txt
Create environment file: copy app/env.example app/.env
Run backend: cd app uvicorn main:app --reload --host 0.0.0.0 --port 8000 http://localhost:8000/docs
- Run Streamlit UI cd cv_app_project pip install -r requirements.txt streamlit run app.py
Open: http://localhost:8501
Future Improvements
Next.js production frontend Authentication & user sessions Deployment pipeline Persistent user profiles Advanced job-match scoring
Note
This project is intended as a demonstration of AI-assisted application architecture and responsible LLM usage, emphasizing transparency and factual correctness.
📄 License
Educational and portfolio use.