Loading projects...
AI-Powered Multi-Provider Resume Analyzer
Structured LLM Qualification Scoring
A highly-structured, multi-provider API that analyzes candidate resumes (text or PDF) against job descriptions to provide a deep, quantitative analysis of skills, qualifications, and interview probability.
PythonFastAPIPydanticInstructorMongoDBpdf2image
Timeline
1 Week
Intensity
Individual
Access
Proprietary
Revision
Mar 2026
The Challenge
Defining the core problem and identified pain points that necessitated this technical intervention.
Hiring managers and recruiters are manually screening hundreds of resumes per job, a process that is slow, subjective, and highly inefficient. They lack a tool to quickly get a data-driven, objective analysis of a candidate's true fit for a role based on the job description.
Technical Solution
The architectural and implementation strategy developed to resolve the challenge.
01
Multi-Provider Architecture
Parallel logic for different LLMs (Gemini, Groq, Cerebras) ensures redundancy and speed-cost tradeoff flexibility, allowing the system to route to the best available provider.
02
Native PDF Processing
Uses Gemini's native PDF processing to extract data directly from `.pdf` files, eliminating the need for external OCR libraries.
03
Structured AI Analysis
The LLM is given a highly-detailed system prompt and forced to return a structured JSON response validated against a comprehensive Pydantic model (`ResumeInsights`), which includes a candidate's extracted profile, quantitative keyword analysis (matched vs. missing), profile match percentage, and a 1-10 Interview Probability Score.
04
Persistent Audit Trail
The final, structured JSON data is saved to MongoDB and returned to the client, providing a persistent audit trail perfect for populating a recruiter's dashboard.
Key Contributions
My specific roles, responsibilities, and the technical value I added to the project lifecycle.
Application Architecture
Architected the entire FastAPI application, including the multi-provider logic to integrate Gemini, Groq (DeepSeek), and Cerebras LLMs.
Pydantic Schema Design
Designed the highly-structured `ResumeInsights` Pydantic model — the core of the project — forcing the AI to return a detailed, multi-part JSON analysis.
System Prompt Engineering
Wrote the complex, 150+ line system prompts that instruct the LLMs on how to perform the analysis and precisely follow the JSON schema.
Multi-Modal Ingestion
Implemented the multi-modal ingestion pipeline, building native PDF processing using Gemini's file-based API (`types.Part.from_bytes`).
Structured Output Integration
Used the `instructor` library to patch the Groq client, enabling structured Pydantic output from the DeepSeek model.
Audit & Persistence Layer
Set up the MongoDB collections to log every job description, resume, and the full structured AI response for auditing and future analysis.
Performance Metrics
Results & Impact
Screening Time Cut
95%
Replaced hours of manual CV review with a structured AI analysis delivered in under 10 seconds per candidate.
Analysis Time
< 10s
Full resume scoring — profile extraction, keyword gap analysis, match percentage, and interview score — in under 10 seconds.
Interview Score
1–10
Every candidate receives a calibrated Interview Probability Score on a 1–10 scale aligned to the specific job description.
LLM Providers
3
Parallel support for Gemini, Groq (DeepSeek), and Cerebras ensures redundancy and speed-cost tradeoff flexibility.
Prompt Lines
150+
Highly detailed 150+ line system prompts enforce strict JSON schema adherence across all supported LLM providers.
Structured Output
100%
Pydantic-validated `ResumeInsights` model guarantees every response is machine-readable — no hallucinated formats.