- Problem Definition
The job application process is fragmented.
A typical candidate has to:
Edit resumes for each job Prepare for interviews Write networking emails Search for relevant roles
These tasks are:
Repetitive Time-consuming Require different tools Objective
Build a single platform that uses Generative AI to streamline the entire workflow.
- Solution Overview
We built Job-Swift AI, a GenAI-powered career assistant that integrates multiple job-seeking tasks into one system.
Core Features: Resume Optimization Mock Interview Simulation Networking Email Generation Job Role Suggestions
- Key Design Philosophy The project is NOT about training AI models It is about engineering workflows around a powerful LLM
So the focus is:
Prompt engineering Input structuring Output post-processing User experience
-
System Architecture High-Level Flow User Input (PDF / Text) ↓ Preprocessing Layer ↓ Prompt Engineering Layer ↓ Gemini 1.5 Pro (LLM API) ↓ Post-processing Layer ↓ UI Output (Streamlit)
-
Input Handling
Example: Resume Upload User uploads a PDF We do NOT send PDF directly to AI We extract text using pdfplumber PDF → Text Extraction → Clean Text
Then: Combine with user input (job description) Inject into prompt template
- Prompt Engineering Layer Instead of: "Improve my resume"
we generate structured prompts like:
Optimize the following resume for this job description.
Resume:
Job Description: Why this matters: Reduces ambiguity Increases relevance Produces consistent outputs
- Model Layer We used: Google Gemini 1.5 Pro
Important clarification: ❌ No internet search ❌ No real-time data ❌ No semantic retrieval (RAG)
✔ Uses pretrained knowledge + prompt context
- Module Breakdown 8.1 Resume Optimizer
Flow:
PDF → Text → Prompt → Gemini → Optimized Resume → PDF
Output: Structured resume Downloadable PDF
8.2 Mock Interview System Flow: Job Role → Prompt → Questions User Answers → Prompt → Feedback
Features:
Multi-question flow Session tracking Structured evaluation: Rating Strengths Improvements
8.3 Networking Email Generator Flow:
User Intent → Prompt → Professional Email
Use cases:
Cold outreach Referral requests LinkedIn messages
8.4 Job Suggestions Flow:
Resume + Preferences → Prompt → Suggested Roles + Companies
These are AI-generated suggestions, NOT real job listings
- Post-Processing Layer
After AI response:
Resume → Converted to PDF using FPDF Interview → Structured display Emails → Editable text box 10. Tech Stack Frontend: Streamlit LLM: Google Gemini 1.5 Pro PDF Handling: pdfplumber, FPDF State Management: Streamlit session state
Key strengths:
Task-specific prompt engineering Modular architecture End-to-end workflow design Handling unstructured inputs (PDFs) Converting AI outputs into usable artifacts
- Limitations
No real job API integration No RAG / embeddings Outputs depend on LLM reliability No automated validation layer API keys not securely managed