Job Copilot — Find, Evaluate & Track Job Applications with AI
A free, no-sign-up job search copilot that runs live on this site. Upload a CV and it becomes a skills profile; search real listings in 11 countries ranked against it; paste any job description for an instant ATS keyword scan and an honest AI verdict (1–5 fit score, cited evidence, red flags, posting-legitimacy check, CV edits and STAR interview stories); then track every application on a drag-and-drop board and draft recruiter emails. A visitor's data never leaves their browser.
- Role: Full Stack & AI Engineer
- Timeline: 2026 - Present
- Team: Solo Project
- Technologies: React, Firebase, Gemini, LLM Agents, Node.js
- Link: https://laxman824.github.io/jobhunt
- Live demo: Try the Job Copilot — Free, no sign-up. Upload your CV or use the sample profile, then search, evaluate and track real jobs.
Problem
Job seekers spray applications because judging fit is slow: reading every JD against your CV, guessing which keywords an ATS filters on, and keeping track of who you emailed. The popular tools split this across paid tiers: keyword scans are capped (Jobscan's free plan allows 5 a month), trackers upsell AI, and most want an account and your data.
Solution
One workspace covering the whole loop. The CV is parsed in the browser (pdf.js) into a structured profile by an LLM. Live search goes through a Firebase Function proxy to Adzuna (keys stay server-side), with Remotive for remote roles, and results are ranked instantly by a local skill-alias matcher before optional AI re-scoring. The evaluator pairs a deterministic, Jobscan-style keyword scan (curated lexicon, plural and alias aware, requirement lines weighted) with an AI judgement modeled on career-ops: five scored dimensions, a 4.0 apply line, requirement-by-requirement evidence quoted from the CV, posting legitimacy kept separate from the score, and a hard rule against inventing experience.
Impact
- Live and public: anyone can use it without an account
- Unlimited ATS keyword scans, computed locally in milliseconds
- Visitor data isolated in the browser; nothing is synced, tracked or sent on their behalf
- Runs on the site's shared LLM chain (Gemini → Groq → NVIDIA) behind per-visitor rate limits and a 7-day response cache
Key features
- CV upload (PDF), paste-in, or a sample profile to explore
- Live job search in 11 countries plus remote-only, ranked against your profile
- Should-I-apply evaluation: 1–5 score, dimension breakdown, cited evidence, red flags, legitimacy check
- Level strategy, top CV changes and STAR interview stories per role
- Drag-and-drop pipeline board (Saved → Drafted → Applied → Callback → Interviewing → Offer) and CSV export
- AI-drafted recruiter emails mapped to the JD, with recruiter-search helpers
- Owner mode: Firestore sync, tracked CV links, one-click Gmail send and a daily follow-up drafter
Tech stack
- Frontend: React, pdf.js, CSS custom properties
- Backend: Firebase Cloud Functions (Node 24), Firestore, Gmail API (owner only)
- Ai: Gemini 2.5 Flash, Groq gpt-oss-120b, NVIDIA Nemotron
- Data: Adzuna API, Remotive API
What K Laxman learned
- Keyword matching is a UX feature, not just a score: showing which required terms are missing is what people act on
- Thinking models spend output budget before the JSON, so structured calls need a generous token ceiling and one silent retry
- A public demo of a personal tool needs a hard data boundary: separate storage scopes, and owner-only side effects
Explore more
- Home — overview, skills and a built-in AI assistant
- Experience — roles at CGI, Think360 AI - CAMS, CAMS Mutual Funds, RAVI - IIT Delhi
- Projects — GenAI, LLM, RAG and full-stack builds
- Education — IIT Delhi, M.Tech & B.Tech Computer Science
- Learn — a working AI engineer's roadmap: LLMs, RAG, agents, evals, production
- GitHub Activity — open-source contributions
- Contact / Hire me