← Back to the portfolio Every score and every line below is real output from the running system, captured from its live API.

ApplyForge

ApplyForge scrapes live job boards, scores every listing against my CV using structured Claude outputs, and writes an application pack for the matches. This page is a static capture of the real database behind it: 21 scored listings, 3 with a generated pack, top score 60. Nothing here is illustrative or rewritten.

Why the scores are low, and why that is the point

A scorer that flatters its owner is useless. This one is built to argue against me: it marks unmet requirements as unmet, writes the gaps in plain language, and refuses to inflate a fit it cannot evidence. The top score on my own database is 60 out of 100. That number is the honest one, and a scorer I could not trust to tell me no would not be worth running.

The ranked list. All 21 scored listings

Sorted by score, exactly as the system ranks them. The qualifications column counts requirements the system judged met against the total it extracted from the ad.

Company names are replaced with a sector descriptor and locations are reduced to a city. These are live vacancies at real recruiters and employers, and publishing my own tool's verdict next to their names would be unfair to them and tells a reader nothing. Every score, title, requirement check, reason and gap below is untouched.

Score	Role	Company, by sector	Location	Fit label	Quals met	Status
60	AI Automation Engineer (LLMs & Agentic AI) - EST Working Hours (Remote)	Remote staffing provider, South Africa	Remote, South Africa	Possible fit	5 of 7	pack generated
60	Business Applications Developer (Microsoft Power Platform & Business Intelligence)	Offshore staffing agency, US client	Remote, South Africa	Possible fit	3 of 7	pack generated
56	AI & Automation Specialist	IT recruiter, Western Cape	Stellenbosch	Possible fit	5 of 8	scored
52	AI & Automation Specialist to Information Technology	IT recruiter, Western Cape	Stellenbosch	Possible fit	4 of 7	scored
45	Power Platform AI Developer	IT services company, Gauteng	Johannesburg	Weak fit	3 of 7	scored
45	Power Platform and AI Developer	IT recruiter	Remote, South Africa	Weak fit	3 of 8	pack generated
38	Microsoft Power Automate AI Specialist	IT services company, Gauteng	Johannesburg	Weak fit	3 of 7	scored
38	Business Intelligence Consultant	Recruiter, undisclosed client	Cape Town	Weak fit	2 of 7	scored
38	Microsoft Power Platform Specialist	Software company	Cape Town, remote	Weak fit	3 of 8	scored
32	Junior AI & Database Developer	Recruiter, engineering and EPCM client	Johannesburg	Weak fit	3 of 7	scored
30	Business Intelligence & Analytics Analyst	Agricultural inputs group	Stellenbosch	Weak fit	2 of 7	scored
28	QA Automation Engineer (Remote)	IT recruiter, online education client	Cape Town, remote	Weak fit	2 of 8	scored
28	Automation Developer	IT recruiter	Johannesburg	Weak fit	3 of 8	scored
22	Legal Engineer	IT recruiter	Cape Town	Weak fit	4 of 8	scored
18	Intermediate Software Developer	Management consultancy	Cape Town	Weak fit	1 of 8	scored
18	Power Platform Solutions Architect	IT recruiter	Cape Town	Weak fit	2 of 7	scored
12	Automation Engineer	Engineering recruiter	Cape Town	Weak fit	2 of 7	scored
11	Automation Engineer	Engineering recruiter	Cape Town	Weak fit	1 of 7	scored
8	Snr Automation Sales Engineer - CPT	Engineering recruiter	Cape Town	Weak fit	1 of 7	scored
8	Snr Automation Sales Engineer - CPT	Engineering recruiter, industrial automation client	Cape Town	Weak fit	1 of 7	scored
7	Automation Engineer	Recruiter, Cape Town	Cape Town	Weak fit	1 of 7	scored

One listing in full. AI Automation Engineer (LLMs & Agentic AI) - EST Working Hours (Remote)

Remote staffing provider, South Africa · Remote, South Africa · scored 60 · Possible fit

Requirements, checked one by one

✓Production AI automations shipped commercially
✓Modern AI coding tools (Claude Code)
✓Orchestration frameworks (n8n, agent workflows)
✓REST API build and integration
✓Docker and deployment practice
✗3-5 years professional software engineering
✗Remote in South Africa, 14:30-23:00 shift

Why it scored what it did

His stack maps almost perfectly onto the ad: production LLM automations, a Claude Code agent rig with subagents and MCP, n8n/Twilio/Retell voice agents, REST APIs and a multi-stage Dockerfile. The blocker is tenure: the ad demands 3-5 years professional software engineering and his commercial track record is months, not years.

The gaps it held against me

Only months of professional engineering experience versus the 3-5 year minimum
Afternoon-to-late-night shift (14:30-23:00) plus SA public holidays; he is currently employed full time and only free after hours for interviews
No LangChain/LangGraph/CrewAI/AutoGen exposure evidenced, only n8n and Claude Code tooling
Mandatory 25 Mbps fibre and backup power not confirmed on CV
No stated salary; ad is a BPO placement so pay may not reach his [pay expectation removed] anchor

The generated cover letter

Redactions are marked in place and nothing else is edited. ApplyForge runs on my own job search data, so its raw output contains my personal contact detail, my pay expectations and real deal values from an employer system. Those and the company name are the only things removed, every removal is shown where it happened, and the rest is the model's output word for word.

I ship production LLM automations, not prototypes. At TrueProp Property Services I built and deployed six live automation systems in my first month, including an AI email intake agent that reads inbound business mail, extracts requestor, department, request and priority, opens numbered tickets and replies automatically, and an AI broker deal pipeline that captured unaided deals of [employer deal values removed] in its first weeks. I also built an OCR classifier that sorts branch scans into eight categories and files them without a human touching them. On the agent and API side, I founded Receva, an AI voice receptionist running on Retell AI, n8n, Twilio and REST APIs, taking real customer calls end to end with booking capture, CRM writeback and follow-up automation. I work daily in Claude Code with custom subagents, hooks, MCP servers and nightly scheduled agents that run research and build tasks unattended, and I use the Claude API with structured outputs in ApplyForge, a Next.js and TypeScript job-matching engine I containerised with a multi-stage Dockerfile down to a 305 MB image, smoke tested end to end. Being straight with you: my professional engineering time is months, not three to five years, so I am shorter on tenure than your minimum and my orchestration work is n8n and Claude Code rather than LangChain or CrewAI. What I bring instead is shipped, monitored, documented production work with measured business impact. The 14:30 to 23:00 EST shift suits me and I am fully remote capable from Cape Town.

One correction, left visible on purpose. This letter was generated before I audited my own numbers, and it says six live systems built. The accurate count is five live in my first month, three built from scratch and two taken over from a departed developer and hardened, plus a sixth built and in business walkthrough. The letter is left exactly as the system produced it, because quietly correcting old output on a page whose whole claim is that the output is real would be the worse choice. The rest of the site carries the corrected figure.

Generated 2026-07-29T15:19:41.321Z. The pack also contains a tailored CV and short application answers. Those are not published here: both are dense with personal contact detail and pay expectations, and redacting them line by line would leave nothing checkable.

What this page does not prove

It does not prove the scores are well calibrated, because there is no hiring outcome to test them against yet. It proves the system ingests real listings, extracts requirements from free text, judges each one against a fixed CV, writes its reasoning down, and produces a finished work product. Calibration needs outcomes, and outcomes need time.