A backend-focused AI system that analyzes a candidate’s resume against a specific Job Description (JD) and returns structured, role-aligned improvement suggestions using a schema-constrained multi-agent LLM workflow plus a deterministic skill-matching layer.

• Overview
• Output Images
• Key Features
• System Architecture
  • Agents
  • Deterministic Skill Matching
  • Orchestration Flow
• Tech Stack
• API Endpoints

The JD-Aware Resume Optimization Engine is a structured multi-agent LLM workflow orchestrated by a deterministic pipeline.

It is designed to:

• Parse resumes and JDs into strict schemas
• Compute explicit skill overlap and gaps deterministically
• Generate gap-driven improvement suggestions with guardrails (no recomputation / no hallucinated skills)
• Return structured JSON suitable for UI rendering

This is not a fully autonomous agent system. It is a multi-stage workflow with:

• Symbolic evaluation (rule-based matching)
• Probabilistic reasoning (LLM-based suggestions)
  kept intentionally separate for control and explainability.

• Resume Parsing Agent

  • Converts extracted resume text into a structured JSON schema
  • Extracts skills, projects, experience summary, certifications, education
  • Uses strict Pydantic schema validation

• JD Parsing Agent

  • Extracts actionable backend requirements and high-signal keywords
  • Removes marketing/boilerplate language
  • Outputs domain focus, keywords, expected experience, etc.

• Deterministic Skill Matching

  • ATS-style overlap scoring (keyword-based)
  • Produces:
    • match percentage
    • matched skills
    • missing skills
  • Ensures transparent, explainable results

• Suggestion Agent

  • Receives: parsed resume schema + parsed JD schema + precomputed gaps
  • Generates targeted improvements only based on identified gaps
  • Guardrails prevent the agent from:
    • recomputing match results
    • inventing new skills not in the gap list

• FastAPI + JWT Authentication

  • Signup/Login protected routes
  • /api/analyze requires Authorization: Bearer <token>

• Frontend

  • React + TailwindCSS + animations
  • Protected routes (only login/signup public)
  • Clean UI and structured result rendering

1. Resume Parsing Agent

   • Input: raw resume text (extracted from PDF/DOC/TXT)
   • Output: strict resume schema (validated)

2. JD Parsing Agent

   • Input: raw JD text
   • Output: strict JD schema (validated)

3. Suggestion Agent

   • Input: resume schema + JD schema + deterministic match results
   • Output: strictly structured suggestions (validated)

Each stage validates output against predefined schemas before passing data forward.

Between parsing and suggestions, the system runs a rule-based matching layer:

• Computes keyword overlap between resume skills and JD requirements
• Identifies missing skills
• Calculates match percentage
• Mimics ATS keyword filtering behavior

This ensures the suggestion layer cannot override the computed skill gap logic.

1. Extract text from resume (PDF/DOC/TXT)
2. Resume Parsing Agent → validated resume JSON
3. JD Parsing Agent → validated JD JSON
4. Deterministic Skill Matching → computed overlap + gaps + % score
5. Suggestion Agent → targeted improvements based on gaps
6. Return structured response to client

Backend

• Python
• FastAPI
• OpenAI LLM APIs
• OpenAI Agents SDK
• Pydantic (schema validation + structured outputs)
• Async orchestration
• JWT Authentication

Frontend

• React (JavaScript + JSX)
• React Router DOM (protected routes)
• TailwindCSS (latest)
• Framer Motion (animations)

Authorization

• /api/analyze requires:
  Authorization: Bearer <JWT_TOKEN>