AI Job Screener

Go monorepo with two services isolated as separate modules:

• services/function: HTTP function responsible for the initial capture of job postings.
• services/worker: batch worker responsible for screening and notification.

Demo

Gravando.2026-08-23.160948.mp4

The repository uses Go Workspaces (go.work, Go 1.25) for joint development, but each service keeps its own go.mod, dependencies and deploy pipeline.

Structure

services/
  function/
    cmd/local/
    internal/http/
    internal/jobs/
    function.go
    go.mod
  worker/
    cmd/worker/
    internal/jobs/
    Dockerfile
    worker.go
    go.mod
deploy/
  function.cloudbuild.yaml
  worker.cloudbuild.yaml
go.work

Local environment

Create a .env file at the repository root:

DATABASE_URL=postgresql://user:password@host/database?sslmode=require
LLM_API_URL=https://api.siliconflow.com/v1/chat/completions
LLM_API_KEY=your-llm-api-key
LLM_MODEL=deepseek-ai/DeepSeek-V3
DISCORD_WEBHOOK_URL=https://discord.com/api/webhooks/your/webhook

Both services try to load .env from the current directory and from the monorepo root.

Development

To build both modules in the workspace:

go build ./services/function/...
go build ./services/worker/...

Function locally

In PowerShell:

$env:FUNCTION_TARGET="Trigger"
$env:LOCAL_ONLY="true"
go run ./services/function/cmd/local

The function receives POST http://localhost:8080/ and forwards internally to /trigger.

Optional payload:

{
  "terms": ["golang", "python backend"]
}

Worker locally

In PowerShell:

go run ./services/worker/cmd/worker

The worker fetches every job with status PENDENTE, processes each item with jitter between requests, calls the configured LLM for a structured review, and notifies Discord with the outcome.

The LLM call follows the POST /v1/chat/completions format, with Authorization: Bearer, Content-Type: application/json, a model field and a single messages item with role: user. The response is read from choices[0].message.content, equivalent to resp.raise_for_status(); resp.json()["choices"][0]["message"]["content"].

Job review and notification

The LLM returns a structured JSON review for each job, containing the fields decisao (APROVADO or REJEITADO), titulo_vaga, empresa, status_processamento and justificativa.

Every processed job — approved or rejected — is posted to the Discord webhook as an embed, color-coded by decision:

• Green (3066993) for APROVADO.
• Red (15158332) for REJEITADO.

The embed includes the job title, company, job ID, decision and application link. Jobs that fail LLM review, LinkedIn detail fetching, or the Discord webhook call are marked as ERRO in the database instead of PROCESSADO.

Service responsibilities

Function

• HTTP stack with Echo and middlewares.
• Endpoint POST /trigger.
• LinkedIn Guest API scraper.
• Initial job persistence in PostgreSQL with deduplication.

Worker

• Batch read of pending jobs.
• Enrichment via LinkedIn's public job detail page.
• Integration with an LLM for structured decision-making.
• Discord webhook notification for every processed job.
• Status update to PROCESSADO or ERRO.

Isolated deploys

Function

gcloud builds submit --config deploy/function.cloudbuild.yaml .

This pipeline deploys the Cloud Function Gen 2 (runtime go125) using only services/function as the service source.

Before the first deploy, create a Secret Manager secret with the PostgreSQL connection string:

printf '%s' 'postgresql://user:password@host/database?sslmode=require' | \
gcloud secrets create DATABASE_URL --data-file=-

If the secret already exists and you want to update its value:

printf '%s' 'postgresql://user:password@host/database?sslmode=require' | \
gcloud secrets versions add DATABASE_URL --data-file=-

The secret name can be changed at deploy time via the _DATABASE_URL_SECRET substitution.

Worker

gcloud builds submit --config deploy/worker.cloudbuild.yaml .

This pipeline builds the worker image from services/worker and publishes it as a separate Cloud Run Job.

Before the first worker deploy, create the required Secret Manager secrets:

printf '%s' 'postgresql://user:password@host/database?sslmode=require' | \
gcloud secrets create DATABASE_URL --data-file=-

printf '%s' 'your-llm-api-key' | \
gcloud secrets create LLM_API_KEY --data-file=-

printf '%s' 'https://discord.com/api/webhooks/...' | \
gcloud secrets create DISCORD_WEBHOOK_URL --data-file=-

If the secrets already exist and you want to update their values:

printf '%s' 'postgresql://user:password@host/database?sslmode=require' | \
gcloud secrets versions add DATABASE_URL --data-file=-

printf '%s' 'your-llm-api-key' | \
gcloud secrets versions add LLM_API_KEY --data-file=-

printf '%s' 'https://discord.com/api/webhooks/...' | \
gcloud secrets versions add DISCORD_WEBHOOK_URL --data-file=-

Secret names can be changed via _DATABASE_URL_SECRET, _LLM_API_KEY_SECRET and _DISCORD_WEBHOOK_SECRET.