proteus

---

⭐ If PROTEUS gave you a smarter way to think about resume optimization — a star helps other engineers find it. Takes 2 seconds.

---

What it does

PROTEUS is a JD-aware resume analyzer that runs a five-agent NVIDIA NIM pipeline to produce consistent, actionable outputs from a single job description and resume.

Output	What you get
Semantic Match Score	Percentage alignment with category breakdown
Gap Analysis	Matched / partial / missing requirements ranked by impact
Bullet Rewrites	JD-aware rewrites with rationale and impact scores
Cover Letter	Tailored letter from the same parsed context
Priority Actions	Ranked steps to improve your application

Every output reads from the same parsed JD context — no contradictions between your score, gaps, rewrites, and cover letter.

How it works

JD ──┐
     ├──→ [Parse] → [Calibrate] → [Map] → [Rewrite] → [Draft] ──→ Results
Resume┘

Step	Agent	Model	Task
01	JD Parser	meta/llama-3.1-70b-instruct	Extract role, requirements, seniority
02	Resume Parser	meta/llama-3.1-70b-instruct	Extract skills, experience, achievements
03	Gap Analyzer	nvidia/nv-embedqa-e5-v5	Semantic similarity scoring
04	Rewriter	meta/llama-3.1-70b-instruct	JD-aware bullet rewrites
05	Cover Letter	meta/llama-3.1-70b-instruct	Tailored letter generation

Models auto-update via GitHub Actions health checks every 6 hours.

Active Models (auto-updated by health check bot)

Role	Model	Last Checked
jd-parser	nvidia/llama-3.3-nemotron-super-49b-v1.5	2026-07-13T09:34:06.578Z
resume-parser	nvidia/gliner-pii	2026-07-13T09:34:06.578Z
gap-analyzer	nvidia/nv-embedqa-e5-v5	2026-07-13T09:34:06.578Z
rewrite-suggester	nvidia/llama-3.3-nemotron-super-49b-v1.5	2026-07-13T09:34:06.578Z
cover-letter	nvidia/llama-3.3-nemotron-super-49b-v1.5	2026-07-13T09:34:06.578Z

Features

Core Analysis

• 3 ways to input a JD: paste, upload, or URL
• 2 ways to input a resume: paste or PDF upload
• Semantic match scoring (embedding-based, not keyword matching)
• Gap analysis ranked by impact with severity badges
• Bullet-level rewrite suggestions with before/after comparison
• Consistent cover letter generated from the same context
• Priority action items ranked by impact

Platform

• Per-user data isolation and analysis history
• Rate limiting (10 analyses/day)
• Streaming API (NDJSON) for real-time pipeline progress
• Mobile-responsive UI
• Dark theme with gold accent design system
• Three auth methods: magic link, Google, GitHub

Reliability

• Automatic model fallback on failure
• JSON retry with temperature escalation (3 attempts)
• 300s timeout on serverless functions
• Health checks every 6 hours via GitHub Actions
• Self-healing model registry (auto-replaces failed models)

Tech Stack

Layer	Technology
Framework	Next.js 16 · App Router · TypeScript
Styling	Tailwind CSS v4 · Dark theme · Geist fonts
AI/ML	NVIDIA NIM · LLaMA 3.1 70B · Embeddings
Database	better-sqlite3 (local) · Turso/libsql (Vercel)
Auth	NextAuth.js v5 · Magic Link · Google · GitHub
Validation	Zod v4
PDF Parsing	unpdf
Deployment	Vercel · GitHub Actions
Rate Limiting	Custom per-user daily limits

API

Endpoint	Method	Description
/api/analyze	POST	Run full pipeline
/api/analyze/stream	POST	Run pipeline with SSE streaming
/api/history	GET	List past runs
/api/history/:id	GET	Get run details
/api/history/:id	DELETE	Delete a run
/api/models	GET	List configured models
/api/health	GET	Health check
/api/health/nim	GET	NIM connectivity test
/api/usage	GET	Daily usage stats

---

Built by Daniel Deshmukh · Mumbai, India