DEV Community

Cover image for JobFit: A 100% Local-First AI Career Copilot That Kills Resume Hallucinations
Samar Shetye
Samar Shetye

Posted on

JobFit: A 100% Local-First AI Career Copilot That Kills Resume Hallucinations

#devchallenge #hf26challenge

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

JobFit is an ethical, local-first AI career application copilot designed to empower job seekers with verifiable, evidence-backed job matching, grounded resume bullet tailoring, and truthful cover letter drafting — operating with zero cloud dependencies and zero external data transmission.

The Problem It Solves

Job seekers and career consultants today face two major hazards with commercial cloud AI tools:

  1. Severe Privacy Breaches: Candidates upload lifetime employment histories, contact info, and addresses to cloud wrappers that log and train on sensitive personal records.
  2. Hallucinated "ATS Scores" & Fabricated Achievements: Commercial matchers invent opaque percentage scores without citing proof. Worse, AI rephrasers frequently fabricate metrics (e.g. turning "built internal portal" into "scaled distributed architecture by 450% generating $12M"), putting candidates at legal risk during background checks.

Who Would Pay For It? (The Side Hustle Angle)

  • Privacy-Conscious Job Seekers & Tech Professionals: Individuals applying for high-trust roles (finance, defense, tech) who cannot leak company IP or personal PII to cloud LLMs.
  • Independent Resume Coaches & Career Consultants: A professional career consultancy running JobFit can offer verified, hallucination-free resume optimization services to clients with zero per-token OpenAI API costs, keeping 100% of the profit margins.

Demo

Youtube Video:

Highlights in the Demo:

  • Multi-Format Ingestion: Ingesting PDF, DOCX, and TXT resumes directly into structured candidate profiles.
  • Evidence-Backed Matching: Transparent mathematical breakdown citing verbatim resume lines for every match.
  • Anti-Fabrication Guardrail: Real-time regex inspection preventing invented numbers or credentials.
  • Application Tracker: Pipeline management with 1-click CSV and Markdown evidence report exports.
  • 1-Click Local Data Purge: Full data sovereignty with instant SQLite and file cleanup.

Code

Github Repository:

GitHub logo Samar-365 / jobfit

JobFit — Local-First AI Job Application Assistant

JobFit is an ethical, local-first AI career application copilot designed to empower candidates with verifiable, evidence-backed job matching, grounded resume bullet tailoring, and truthful cover letter drafting — operating with zero cloud dependencies and zero external data transmission.

The Problem with Cloud Career Tools

  1. Severe Privacy Hazards: Candidates upload their lifetime work histories, contact numbers, addresses, and compensation notes to third-party cloud LLM wrappers that log and train on sensitive personal records.
  2. Hallucinated "ATS Scores": Commercial resume matchers use black-box scoring algorithms that invent scores (e.g., "78% ATS Match") without explaining what is missing or citing proof.
  3. Fabricated Credentials: Cloud AI rephrasers frequently invent numbers (e.g., changing "built internal portal" into "scaled distributed architecture by 450% generating $12M"), putting candidates at legal and ethical risk during background checks.

The JobFit Solution

JobFit solves these problems by…

View on GitHub

How I Built It

JobFit is engineered as an entirely self-contained workstation application:

The Open-Source AI Engine

  • Google Gemma 2 (2B): Quantized (Q4_K_M, ~1.6 GB) running locally via Ollama. It delivers sub-second to 15s inference directly on consumer GPUs (NVIDIA RTX 4050 Laptop) or modern CPUs with zero cloud API keys.
  • Fenced XML Prompt Boundaries: User-supplied resumes and job postings are strictly isolated within <resume_document> and <job_posting> XML delimiters to prevent prompt injection and adversarial system overrides.
  • Verbatim Citation Extraction: Gemma is prompted to act as an objective auditor, required to extract exact quotes for matches or classify criteria as NOT_MENTIONED rather than hallucinating shortcomings.

The Application Stack

  • Backend: Python 3.11, FastAPI, SQLite, SQLAlchemy 2.0, pypdf, python-docx.
  • Frontend: React 18, TypeScript, Vite 6, styled with a custom Pastel 3D Studio Design System (tactile 3D cards, porcelain off-white canvas #F8F7F4, soft pastel accents).
  • Deterministic Scoring Engine: Replaces black-box heuristics with an open formula: $$\text{Score} = \left(\frac{\text{Weighted Verified Points}}{\text{Total Criteria}}\right) \times 100$$
  • Programmatic Guardrail Validator: Inspects LLM bullet suggestions against the candidate's original text to mathematically reject any hallucinated metrics or percentages.

Why Does Open Innovation Matter?

Open-weight models like Google Gemma 2 make applications like JobFit possible in ways closed commercial APIs (OpenAI, Anthropic) cannot:

  1. Zero Data Transmission: Career data contains the most intimate summary of a person's life. Open-source local inference is the only guarantee that a candidate's personal data never leaves their physical machine.
  2. Predictable Unit Economics: A freelance career consultant can run millions of evaluation tokens locally through Ollama with $0 in API bills, turning an AI side hustle into a sustainable, high-margin business.
  3. No Vendor Lock-In or Sudden Deprecations: The entire stack is self-hosted, reproducible, and verifiable.

My First User

To test JobFit in the real world, I shared the build with a fellow software engineering friend who has been actively applying for roles in the current tough tech market.

Their Experience & Feedback:

  • What they tested: They uploaded their real resume (Alex Rivera archetype) and ran it against a Senior Fullstack React/Python job posting.
  • What impressed them most:
    1. Honest Citations: They were amazed that every single requirement match cited the exact verbatim sentence from their resume instead of giving an arbitrary black-box score like commercial tools.
    2. No Hallucinated Lies: In their words: "ChatGPT always turns my simple API projects into 'scaled enterprise infrastructure generating $5M'. JobFit's tailoring kept my numbers accurate so I won't get embarrassed in a technical interview."
    3. Privacy Peace of Mind: They loved knowing their phone number, email, and personal background never touched a cloud server.

Would they pay?

Yes. They mentioned they currently pay $20/month for a popular cloud resume scanner that hallucinates constantly. They said they would readily pay a $15–$25 one-time lifetime license (or $10/month) for JobFit as an offline desktop application because of the zero-cloud privacy guarantee and honest evidence matching.

What they asked for next:

  1. One-click PDF export for the tailored resume.
  2. A direct URL scraper to paste a LinkedIn or Indeed job link instead of copying and pasting the job text.

Prize Categories

Best Use of Gemma ($200)

JobFit is built natively around Google Gemma 2 (2B) (Q4_K_M quantization) running 100% locally on the candidate's workstation via Ollama:

  • Local Sovereignty: Runs on localhost:11434 with zero external cloud dependencies or telemetry, ensuring sensitive personal career history never leaves the user's laptop.
  • Evidence-Backed Fact Extraction: Gemma acts as an objective evaluator, extracting exact verbatim quotes from submitted resumes to verify qualifications rather than inventing arbitrary "ATS match scores".
  • Prompt Injection Defense: Gemma's inference is encapsulated inside XML-delimited prompt boundaries (<resume_document> and <job_posting>), successfully neutralizing adversarial prompt injection attacks.

Top comments (0)