{"slug": "i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on", "title": "I Built a Job Match Agent with Hermes Agent, And It Texts Me the Results on WhatsApp", "summary": "A full-stack and AI engineer built a WhatsApp-integrated job match agent using the Hermes Agent framework that autonomously analyzes job postings against a user's CV and returns a scored fit analysis and tailored cover letter within two minutes. The system, called Job Match Agent, runs on a Windows machine and uses a custom Hermes skill written in markdown with YAML frontmatter to plan and execute multi-step reasoning, including fetching job descriptions, extracting CV data, and generating responses without a hardcoded pipeline. In tests against two live job postings, the agent produced scores of 64/100 and 84/100, with detailed breakdowns of required skills, experience, domain match, and verdicts.", "body_md": "*This is a submission for the Hermes Agent Challenge*\n\nJob hunting is tedious. You find a posting, you open your CV, you squint at both for 20 minutes trying to figure out if you're actually a good fit, then you spend another hour writing a cover letter that probably sounds like every other cover letter.\n\nI'm a full-stack and AI engineer currently on the market. I thought: what if I could just send a job URL to my phone and get back a scored analysis and a tailored cover letter while I'm on the bus?\n\nThat's what I built with Hermes Agent.\n\n**Job Match Agent** is a custom Hermes skill that:\n\nThe whole thing runs on my Windows machine. I talk to it from my phone. One message in, full analysis back.\n\nFlow:Type \"Job match this: [URL]\" on WhatsApp → Hermes fetches the job, reads your CV, scores fit, writes cover letter → full response back on your phone in ~2 minutes.\n\nGitHub: [hermes-job-match-agent](https://github.com/nnoromiv/job-match)\n\nThis is where it gets interesting. I didn't write a script that does these steps mechanically. I wrote a **skill** a markdown file that teaches Hermes Agent what to do and Hermes handles all the planning and reasoning itself.\n\nWhen I send a job URL, here's what Hermes actually does autonomously:\n\n`job-match`\n\nskill from its skills directory`read_cv.py`\n\nto extract my CV text from the PDF on disk`fetch_job.py`\n\nto scrape the job description from the URLThat's real multi-step agentic behavior; **planning, tool execution, reasoning, output**, not a hardcoded pipeline. Hermes decides the order of operations, handles failures gracefully, and formats the final response for WhatsApp automatically.\n\nThe **memory system** is what makes the WhatsApp UX smooth. I told Hermes my CV path once:\n\n\"Please remember my CV is always at: C:\\Users...\\Christian_Nnorom.pdf\"\n\nHermes saved it. Now from my phone I just send:\n\n\"Job match this: [URL]\"\n\nNo repeating myself. No attaching files. Hermes already knows.\n\nThe two Python scripts are focused tools. All the intelligence the planning, scoring rubric, and cover letter logic, lives in the skill's instructions, which Hermes reads and follows.\n\nHermes skills are markdown files with YAML frontmatter. No plugin registration, no framework boilerplate. Drop a `SKILL.md`\n\ninto the skills folder and Hermes picks it up automatically.\n\n```\n---\nname: job-match\ndescription: \"Analyze a job posting against a CV/resume to score fit, explain reasoning, and generate a tailored cover letter.\"\nversion: 1.0.0\nauthor: Christian Nnorom\nplatforms: [linux, macos, windows]\nmetadata:\n  hermes:\n    tags: [Jobs, Career, CV, Resume, Cover Letter, Recruitment]\n---\n```\n\nThe skill defines the four-step workflow in plain English, with scoring criteria, output format templates, and fallback handling. Hermes reads this and executes it — including deciding when to run which script and how to handle edge cases like a URL that requires JavaScript rendering.\n\nI tested it against two live job postings on the same day I built it.\n\n**Role 1 — KTP Associate, AI & Computer Vision (UWE Bristol / Foster + Freeman)**\n\n```\nOverall Score: 64/100\n\nRequired Skills: 20/40 — Strong ML/CV experience but no PhD (mandatory)\nExperience Level: 12/20 — 3+ years industry experience, MSc distinction\nDomain Match: 14/20 — Applied AI/CV work relevant, forensic context less so\nNice-to-haves: 18/20 — Docker, AWS, MLOps, full-stack all present\n\nVerdict: Apply. Lead with production AI pipeline experience. Address the\nPhD gap directly by highlighting MSc distinction and immediate readiness.\n```\n\n**Role 2 — AI Platform Engineer, KTP Associate (Aston University / Modular Data Ltd)**\n\n```\nOverall Score: 84/100\n\nRequired Skills: 32/40 — Strong match on cloud/microservices, MLOps, Python,\n  ethical AI (SHAP/LIME). Gaps: knowledge graphs, Go/Java.\nExperience Level: 18/20 — MSc Distinction + 3+ years production AI systems\nDomain Match: 16/20 — Modular AI platform work aligns directly\nNice-to-haves: 18/20 — Open-source, AWS cert, multi-cloud, ethical AI all present\n\nVerdict: Strong fit. Emphasise modular platform delivery and MLOps expertise.\n```\n\nTwo roles, two honest scores — 64 and 84. The agent called out the PhD gap on the first role rather than inflating the score. That's exactly what you want from a tool like this.\n\nEach analysis came with a full cover letter — 248 words, keywords matched from the job description, no generic openers.\n\n** read_cv.py** — tries\n\n`pdfplumber`\n\nfirst, falls back to `pypdf`\n\n, prints extracted text with a word/line count header. Fails loudly with actionable errors if extraction fails.** fetch_job.py** — uses\n\n`httpx`\n\nwith a real browser User-Agent, then `beautifulsoup4`\n\nto extract the job container using a prioritised list of selectors covering LinkedIn, Greenhouse, Lever, Workday, and generic semantic tags. Falls back gracefully with instructions to paste text directly if the page requires JavaScript.\n\n```\npip install pdfplumber httpx beautifulsoup4\n```\n\nFull code at: [github.com/nnoromiv/job-match](https://github.com/nnoromiv/job-match)\n\n**Skills are surprisingly powerful.** Writing the workflow in markdown felt almost too simple, but it works because Hermes actually reads and follows the instructions rather than vaguely gesturing at them.\n\n**The memory system is the difference between a demo and a real tool.** Storing the CV path once and never repeating it makes the WhatsApp UX feel like a product, not a hack.\n\n**Honest agentic output requires honest prompting.** The scoring rubric explicitly says \"be honest in the gap analysis a useful score requires accurate assessment.\" Without that, agents flatter. With it, you get a 64 that's actually useful.\n\nThe foundation is solid. Hermes Agent handles the orchestration; extending the skill is all it takes.\n\n`job-match`\n\nfolder into your Hermes skills directory`pip install pdfplumber httpx beautifulsoup4`\n\n*Built in one day for the Hermes Agent Challenge. Running on a Windows VPS, reachable from WhatsApp.*", "url": "https://wpnews.pro/news/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on", "canonical_source": "https://dev.to/nnoromiv/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on-whatsapp-281o", "published_at": "2026-05-30 12:46:06+00:00", "updated_at": "2026-05-30 13:13:01.948838+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "generative-ai", "natural-language-processing", "large-language-models"], "entities": ["Hermes Agent", "Job Match Agent", "WhatsApp", "GitHub", "Hermes"], "alternates": {"html": "https://wpnews.pro/news/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on", "markdown": "https://wpnews.pro/news/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on.md", "text": "https://wpnews.pro/news/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on.txt", "jsonld": "https://wpnews.pro/news/i-built-a-job-match-agent-with-hermes-agent-and-it-texts-me-the-results-on.jsonld"}}