An AI HR Assistant your employees can chat with about their leave, payslips, and policies — live now. Plus AI-powered HR Analytics that flag attrition risk and headcount drift before they hit the boardroom. No abstract hype, just two things HR actually does, faster. Match AI and JD Generator follow in Q3 2026.
Match AI for hiring, HR Assistant for employees, JD Generator for managers.
Match AI uses a 100-point algorithm. Same input = same score, every time.
DOLE handbook, RA 11210, RA 11058, BIR brackets. Filipino HR aware.
We did not bolt a chatbot onto everything. Each tool is shaped to one HR workflow, with explicit data boundaries and honest limits.
Deterministic 100-point algorithm. 6 dimensions: role relevance, skills, experience, education, location, salary alignment. Same input always gives the same score. No black box.
See how scoring worksEmbedded in the Employee Self-Service portal. Answers using only the employee's own data. PII-minimized. PH labor-law-aware.
Explore the assistantOutput: markdown description, suggested responsibilities, must-have and nice-to-have qualifications, compensation band guess. Save as template or paste into a posting wizard.
Try the JD generatorEvery match score is a sum of six measurable signals. The breakdown is returned with the score, so you can explain it to a candidate, a hiring manager, or yourself.
If a vendor pitch sounds magical, it usually means someone is hiding a tradeoff. Here is the list of things our AI explicitly will not do, and why.
Match AI scores, it does not decide. A 32/100 candidate still appears in the pipeline. A human recruiter must move every candidate forward or reject them. Final decisions are not delegated to the model.
An employee asking the chat "how much leave does Anna have?"will be told no. The system prompt is scoped to the requesting user's own row, balances, and payslip summary — and only those.
It produces draft markdown plus a salary band guess. You read it, you edit it, you publish it.Generated content is marked as draft until a human approves — never auto-pushed to JobStreet, LinkedIn, or your careers site.
Those fields are not inputs to the Match AI rule set. Even if a CV mentions them, they are stripped before scoring. Only job-relevant signals are scored: role, skills, experience, education, location, salary.
If an employee asks the AI Assistant "how many years until I get retirement pay?"and we don't have the data, it will say so — not hallucinate. Every answer is grounded in a record we can show you.
Use this as a quick map before you click through to a deeper page. Where the AI lives, who uses it, what model is behind it, and where the data goes.
| Tool | Lives in | Used by | Engine | Output |
|---|---|---|---|---|
| WORKSPHR Match AI | Recruitment & ATS, candidate cards, talent search | Recruiters, hiring managers | Deterministic 100-pt rule engine, no LLM | Score 0–100 + per-dimension insight, sortable shortlist |
| AI HR Assistant | Employee Self-Service portal, mobile app | Employees | LLM GPT-4o-mini, scoped to user data | Chat answer + cited record (e.g. payslip ID) |
| AI JD Generator | Recruitment — Job Description Generator page | HR & hiring managers | LLM GPT-4o-mini + PH context | Markdown JD, responsibilities, must-have / nice-to-have, salary band |
Most "AI matching" tools today are vibes in a black box. Match AI is the opposite, by design. Here is what you get when scoring is rule-based instead of learned.
Every score returns six per-dimension insights. "5 years experience meets 3+ year requirement" beats "the model gave you 87". If a candidate asks why they were ranked lower, your recruiter has a real answer.
Learned models inherit whatever bias was in their training set, including hiring patterns from an era you would not want to repeat. Match AI has zero training data. Bias has to be put in by you, in the rules, where you can see and change it.
Every match score, every chat message, every JD generation is logged with timestamp, input, output, and model version. Same input gives the same score, forever, so you can re-run a hiring decision a year later and prove the math.
Book a 30-minute demo. We'll score a sample of your candidate pool live, walk through a real chat session in ESS, and generate a JD for one of your open roles — on-call, no slides.