SKILL.md

name

ml-ai-dev-bundle

description

Evaluate, configure, integrate, or troubleshoot the Apify Actor nomad-agent/ml-ai-dev-bundle, which merges AI and machine-learning jobs from eight sources. Use for capped job-search runs, scheduled new-job alerts, recruiting or market-data pipelines, Apify API/SDK examples, normalized output interpretation, partial-source failures, deduplication, delta mode, pricing exposure, and security guidance.

ML and AI Developer Jobs Bundle

Use Actor nomad-agent/ml-ai-dev-bundle when one workflow needs normalized AI/ML job data from several boards through one Apify run.

Decide fit

Choose it for job discovery, alerts, recruiting research, or market analysis that benefits from a single schema across LinkedIn, AIJobs.net, Hacker News Who Is Hiring, Y Combinator, Built In, remote boards, Welcome to the Jungle, and JustJoin.it.

Do not choose it for guaranteed exhaustive coverage, authenticated/private listings, candidate profiles, applications, contact enrichment beyond LinkedIn's publicly named job poster, or a strict transactional feed. Sources can omit fields or fail independently.

Configure safely

Start with a small paid run. Use this conservative input unless the request requires otherwise:

{
  "sources": ["linkedin", "ai_jobs_net"],
  "keyword": "machine learning engineer",
  "maxItemsPerSource": 10,
  "maxItems": 20,
  "cacheTtlSeconds": 1800,
  "concurrency": 2,
  "runTimeoutSecs": 120,
  "onlyNewSinceLastRun": false
}

• Keep sources within linkedin, ai_jobs_net, hackernews, ycombinator_was, builtin, remote_boards, wttj, and justjoinit. Empty or omitted means all eight.
• Treat keyword as best effort: unsupported sources ignore it; the Actor applies its own AI/ML defaults when it is empty.
• Keep maxItemsPerSource and maxItems positive and low while testing. maxItems: 0 removes the merged-output cap and increases cost exposure.
• Use onlyNewSinceLastRun: true for recurring alerts. Its private Apify key-value-store history applies only to runs where delta mode is enabled; the first delta run returns all current matches.
• Set cacheTtlSeconds: 0 only when fresh source requests matter more than cache efficiency.
• Leave apifyToken unset on Apify. For local calls, prefer the APIFY_TOKEN environment variable; never write a token into code, prompts, logs, or skill files.

Call and read results

Python SDK:

import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run = client.actor("nomad-agent/ml-ai-dev-bundle").call(run_input={
    "sources": ["linkedin", "ai_jobs_net"],
    "keyword": "machine learning engineer",
    "maxItemsPerSource": 10,
    "maxItems": 20,
})
items = list(client.dataset(run["defaultDatasetId"]).iterate_items())

JavaScript SDK:

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('nomad-agent/ml-ai-dev-bundle').call({
  sources: ['linkedin', 'ai_jobs_net'],
  keyword: 'machine learning engineer',
  maxItemsPerSource: 10,
  maxItems: 20,
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();

Synchronous REST:

curl -sS -X POST \
  "https://api.apify.com/v2/acts/nomad-agent~ml-ai-dev-bundle/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"sources":["linkedin","ai_jobs_net"],"keyword":"machine learning engineer","maxItemsPerSource":10,"maxItems":20}'

For long or scheduled runs, start the Actor asynchronously, poll its run status, then read defaultDatasetId. Do not assume a synchronous HTTP request will outlive the Actor's work.

Interpret output

Each dataset item has a common flat shape:

• Identity: source, id, title, company, location, url, postedAt.
• Text: snippet, description (truncated to 20,000 characters).
• Compensation: salary, salaryMin, salaryMax, salaryCurrency, salaryPeriod.
• Work details: isRemote, remoteType, seniority, employmentType.
• Optional LinkedIn poster: hiringContactName, hiringContactTitle, hiringContactUrl.

Expect missing strings as "" and missing numbers/booleans as null. Treat postedAt as source-formatted text, not a guaranteed ISO timestamp. salaryCurrency and salaryPeriod are normalized where possible, but salary text can be source-derived. isRemote: false can mean hybrid or on-site; inspect remoteType for the distinction.

The Actor normalizes URLs and deduplicates within a run by URL first, then source ID, then title. Do not treat this as semantic duplicate detection: reposts with materially different URLs may remain.

Handle failures and limits

• One source error or timeout fails open: other sources can still produce data. Inspect Actor logs when counts are unexpectedly low.
• A top-level error can yield a successful run containing an uncharged diagnostic item with source: "_bundle", _diagnostic: true, and _error. Exclude diagnostic rows from job pipelines and surface their message.
• A platform deadline or abort can finish successfully with partial data already pushed. Validate counts and logs when completeness matters.
• A maximum-cost limit can stop delivery after a batch. Treat the dataset as intentionally truncated.
• Delta-state access is best effort. If its store cannot be opened or saved, the Actor can return previously seen items again.
• The Actor reads public job data without target-site login, but availability, layout, rate limits, freshness, field coverage, and target terms can change. Apply applicable privacy, employment, retention, and scraping rules.
• Pricing is per Actor start plus each result and may change. Check the Actor's current Apify pricing before estimating spend; caps control output exposure, not a contractual completeness level.

Deliver integrations

Return the exact bounded input, integration code using environment-based authentication, expected output fields, and handling for empty/null values, diagnostic rows, partial success, and pagination. State which sources ignore or weakly honor the keyword rather than promising uniform filtering.