Multi-agent AI pipeline · Claude + Gemini

Land interviews
on autopilot.

A multi-agent resume pipeline that extracts JD tokens, rewrites your resume, and validates ATS scores — targeting 95%+ on Greenhouse and 98%+ on Workday.

Claude Sonnet 4.6Gemini 3.5 FlashATS Scoring LoopDOCX ExportWorkday-Strict
Why Relay

Built for scale and precision.

Six specialized agents, optimized for every major ATS system, powered by best-in-class infrastructure.

6
Specialized agents
Claude + Gemini
98%
ATS target
Ultra-strict parsing
< 3m
Pipeline runtime
End-to-end
98%
Exact token match
Plural & tense aware

Optimized For The ATS Systems That Matter

Greenhouse≥ 95%

Exact token match

Workday≥ 98%

Ultra-strict parsing

LeverCompat.

Standard headers

iCIMSCompat.

Plain DOCX

Claude
Gemini
Next.js
Supabase
Tailwind CSS
FastAPI
Python
WebSockets
Claude
Gemini
Next.js
Supabase
Tailwind CSS
FastAPI
Python
WebSockets
The Pipeline

Six agents. Under 3 minutes.

Watch the relay. Each agent hands off to the next.

JD Overview
Company · role · seniority
Gemini
Token Extract
Keywords & skills
Claude
Resume Writer
Per-job rewrite
Claude
ATS Validator
Score & retry
Claude
Resume Review
Second opinion
Gemini
DOCX Export
Supabase storage
Claude
Phase 0: JD Parser + Token Extract run in parallel
Claude Sonnet 4.6
Gemini 3.5 Flash
Typical runtime: 1–3 minutes
The Product

A dashboard built for throughput.

Submit jobs, watch the live pipeline, grab your DOCX. Everything in one place.

9:41
📶🔋
R
Relay
Dash
Recent Jobs
Backend Engineer
99%
Anthropic
Frontend Engineer
99%
Amazon
ML Engineer
99%
OpenAI
Features

Every agent does one thing perfectly.

Six specialized agents in a coordinated pipeline.
JD Parsing, Token Extraction, Writing, Validation, Review, Export.

Intelligent JD Parsing

Gemini extracts company, role, seniority, and a structured overview from raw JD text — scraped or pasted.

Exact Token Extraction

Claude mines must-have keywords, skills, and phrases from the JD — preserving exact plural/tense to beat strict ATS parsers.

Per-Job Resume Writing

Claude rewrites your resume for the target role. Forbidden verbs are blocked. Bullet realism rules enforced.

ATS Validation Loop

Claude scores the generated resume against the JD token set. Retries with feedback until target threshold is hit.

Reviewer

Second-opinion pass from Gemini catches tone issues, redundancy, and weak phrasings after the main write.

Clean DOCX Export

Delivered as an ATS-safe DOCX stored in Supabase. Ready to upload. No parsing surprises.

Why It Works

Precision engineering, not prompt hacks.

Every design decision is deliberate. No magic. Just careful orchestration of the best models for each job.

Multi-model

Right model for every phase.

Claude Sonnet 4.6 handles writing, token extraction, and validation — where rule-adherence matters most. Gemini 3.5 Flash handles JD parsing and post-write review — where breadth of understanding wins.

Claude Sonnet 4.6
3 stages
Gemini 3.5 Flash
2 stages
Pipeline timeline
JD Parser
Token Extract
Resume Writer
ATS Validator
Resume Review
Output Writer
JD Parser + Token Extract run in parallel — Phase 0
Speed
< 3 min end-to-end.
< 3m
Accuracy
Exact-token match.
Plural
Tense
Header
Security
Hardened against injection.
4-Layer
Export
ATS-safe DOCX, every time.
DOCXPlainUnicode
ATS Targets

Aim for the score the ATS actually reads.

Different ATS systems score differently. Relay tunes its validation loop to each platform's strictness.

Greenhouse

greenhouse.io

≥ 95%

Target

Standard exact token matching

  • Exact keyword match on JD tokens
  • Standard section headers enforced
  • Bullet realism & verb seniority
  • Retries until threshold achieved

Workday

myworkdayjobs.com

≥ 98%

Target

Ultra-strict parsing

  • Exact plural/singular matching
  • Strict verb tense enforcement
  • Canonical date formatting
  • Workday-only section headers
Security First

Built to resist
prompt injection.

Every piece of user-supplied data passes through a four-layer defense before it ever reaches an LLM. No single failure can hijack the pipeline.

01
Pydantic Input Validation
Length caps, structural checks, and suspicious-keyword warnings at the API boundary.
02
XML Entity Escaping
All user content is escaped before f-string injection into prompts.
03
USER_DATA Boundaries
Tagged markers tell Claude to treat the block as data, not instructions.
04
Constrained Output Format
System prompt enforces strict output tags, limiting injection blast radius.
Validation
Escape
Boundaries
Output
5
Attack vectors tested
JWT
Every endpoint
RLS
Row-level isolation
How It Works

From JD to ATS-ready in under 3 minutes.

01

Paste the JD

Drop a job description URL or paste the text. Greenhouse and Workday auto-detected from the URL.

02

Agents go to work

JD overview → token extraction → resume writing → ATS validation → Gemini review. All in parallel where possible.

03

Download your DOCX

Clean, ATS-safe Word doc delivered to your dashboard. Hit apply with confidence.

Testimonials

Built for people who apply to lots of jobs.

I was spending hours per JD rewriting bullets. Now I paste the link, wait a couple minutes, and upload the DOCX. The Workday scores are wild.

P
Priya N.
ML Engineer

The ATS validation loop is the cheat code. It retries until it hits the threshold, so every resume that lands in my hands is already optimized.

M
Marcus T.
Data Scientist

I like that it uses Claude for writing and Gemini for review. Two perspectives on every resume means way fewer weird phrasings slip through.

J
Jordan L.
Backend Engineer
Pricing

Power your job search.

Relay is a parallel job-application operating system. Pay for throughput and control — not storage.

Free
$0forever

Prove the value before committing.

  • 10 resumes / month
  • 1 parallel run at a time
  • Claude Haiku + Gemini 3.5 Flash
  • PDF export
  • Basic ATS optimization
  • Organized by company & role
  • Community support
Get Started Free
Most Popular
Standard
$15/ month

The full product. For active job seekers.

  • 200 resumes / month
  • 4 parallel runs
  • Claude Sonnet + Gemini 3.5 Flash
  • Batch 20 jobs at once
  • 500 spaces
  • 15 GB storage
  • PDF + DOCX export
  • Advanced ATS reruns
  • Faster queue
  • Prompt view & guided chat
  • Version compare
  • 2 resume profiles
Get Standard
Pro
$39/ month

Maximum throughput and control.

  • 300 resumes / month
  • 10 parallel runs
  • Claude Opus + Sonnet + Haiku + Gemini 3.5 Flash
  • Batch 50 jobs at once
  • Unlimited spaces
  • Unlimited storage
  • PDF + DOCX export
  • Advanced ATS reruns
  • Priority queue
  • Prompt editing & versioning
  • Prompt view & guided chat
  • Version compare
  • 5 resume profiles
Get Pro

No credit card required for Free. Cancel Standard or Pro anytime.

FAQ

Questions, answered.

How is this different from ChatGPT rewriting my resume?
ChatGPT gives you one shot with no validation. Relay runs a dedicated ATS validator after every write and retries until the score hits the target. It also enforces rules like forbidden verbs, canonical date formats, and exact plural/tense matching — things a general chatbot won't track. And since one run takes 1–3 minutes, Relay's parallel execution means you can queue up multiple applications while the first one finishes.
Which LLMs does it use?
Claude Sonnet 4.6 handles token extraction, resume writing, and ATS validation. Gemini 3.5 Flash handles JD parsing and post-write review. Each model is used where it performs best — Claude for rule-strict writing, Gemini for broad understanding.
What's the difference between Greenhouse and Workday modes?
Greenhouse mode targets ≥95% with standard exact-token matching. Workday mode targets ≥98% and is ultra-strict about exact plural/singular forms, verb tense, and only accepts standard section headers. Workday parsers are unforgiving, so the bar is higher.
Is my data safe?
Yes. Every API endpoint requires a JWT. User data goes through four layers of defense before reaching any LLM: Pydantic validation, XML escaping, USER_DATA boundary markers, and a constrained output format. Row-Level Security isolates your data from other users.
Do you store my resume?
Your profile data and generated DOCX files are stored in Supabase (your private bucket), accessible only to your authenticated session. You can delete everything from your profile page at any time. Older outputs are archived, not deleted — your history is always available.
Can I customize the prompts?
Standard users can view prompt templates and use guided prompt chat. Pro users can fully edit every prompt template, access the full prompt history, and version-restore previous prompts.
Why pay for Relay when I can use ChatGPT?
ChatGPT is a text tool. Relay is a workflow system. Relay handles ATS scoring, retry loops, multi-step pipelines, company/role organization, parallel execution, and persistent output storage. You spend less time prompting and waiting — Relay runs in the background while you keep applying.

Ready to stop
writing resumes?

Let six specialized agents do the heavy lifting. You focus on the interview.