Skip to main content
How it works

One pipeline, from job posting to curriculum change

Every stage produces structured output the next stage consumes. Nothing in the chain is a black box: each figure can be traced back to the postings that produced it.

01

Ingest labour-market signals

Job postings and employer-declared requirements enter the platform.

  • Manual entry, CSV import, employer submission, or a future API feed — all behind one provider interface
  • Content hashing suppresses the same posting arriving from two sources
  • Location and sector are resolved to canonical records, with unrecognised values reported rather than silently dropped
  • Postings dated in the future are clamped instead of discarded, because that is a common feed defect
02

Extract skills from free text

NLP turns unstructured hiring prose into countable skill signals.

  • The description is tokenised and swept for n-grams up to four words, longest match first
  • Each candidate phrase is resolved against the skill taxonomy and its alias index
  • Proficiency is inferred from surrounding language ("strong", "working knowledge", "exposure to")
  • Experience, education and certifications are parsed by rule
  • Every match carries a confidence score; anything below the threshold is flagged for human review
03

Normalise onto canonical skills

Naming variants collapse to one skill before anything is counted.

  • "MS SQL", "SQL Server" and "Microsoft SQL Server" all resolve to one canonical row
  • Diacritics, punctuation, plurals and filler words are handled during normalisation
  • Multi-word containment lets "SQL Server database administration" still resolve, with confidence reduced for the unexplained remainder
  • Unrecognised technology-shaped tokens are surfaced so an administrator can promote genuinely new skills into the taxonomy
04

Score demand

A composite 0–100 demand index per skill, per district, per sector.

  • Five weighted components: frequency, posting volume, employer validation, recency and sector growth
  • Volume is log-damped so large districts do not swamp small ones
  • Employer-declared demand is weighted above inferred posting signals
  • Confidence is driven by sample size and employer corroboration, and is reported alongside every score
  • All weights live in configuration, so the model is tunable per deployment
05

Detect drift

How the required skill mix is changing over time.

  • Frequency snapshots are captured per period and stored immutably
  • Consecutive periods are diffed to classify each skill as emerging, growing, stable or declining
  • Emerging status requires both a large relative jump and enough absolute presence — tripling from 0.1% to 0.3% is not an emerging skill
  • A skill is never called obsolete on one period's evidence; sustained decline across multiple periods is required
06

Align curriculum

The Curriculum Twin is diffed against the demand graph.

  • Each course is modelled as structured data: modules, topics, hours, practical split, skills taught and proficiency reached
  • Two coverage figures are produced: plain coverage, and coverage weighted by demand score
  • The weighted figure is the honest one — covering ten low-demand skills does not offset missing the one every employer asks for
  • Missing, emerging-missing, overrepresented and genuinely obsolete content are reported separately
07

Reconcile district supply and demand

Demand index against training supply index, district by district.

  • Demand index from open vacancies and active postings
  • Supply index from annual seat capacity and available trainer capacity
  • Gap score is the difference, classified by configurable severity thresholds
  • Per-skill shortages are ranked by demand weighted against local trainers, courses and seats
08

Generate explainable recommendations

Specific actions, each carrying the evidence that produced it.

  • Targeted at the institute, the planner, the employer or the individual student
  • Every recommendation states the numbers behind it — posting counts, share, current coverage, employer validation
  • Priority and impact are derived from demand score and gap severity
  • Recommendations are keyed so regeneration updates rather than duplicates, and a planner's decisions survive the next run
09

Close the loop with employers

Validation feeds back into confidence and priority.

  • Employers confirm or reject the demand the system detected in their district
  • Validated signals raise the confidence of every recommendation that depends on them
  • Rejected signals are suppressed from scoring
  • The result is a system that gets more accurate the more it is used

Walk the loop yourself

Sign in with a demonstration account and follow the same journey end to end: district demand → curriculum gap → employer validation → student learning path.

Explore the Platform