Tekzad AI logoTekzad AIIndependent engineering sample

UNIGUIDE CONTEXT / AGENTS & RETRIEVAL

A learning path.
With the reasoning attached.

A prerequisite-aware planner, backed by a small searchable evidence catalogue. See what fits your time, what depends on what, and where an educator should step in.

All courses and progress are synthetic. No learner records, live model, LMS or academic decisions are connected.

Build a plan ↘

01 / LEARNER CONSTRAINTS

Start with what you can verify.

-Remaining learning hours
-Weeks at this budget
-Current target skills covered

Prerequisite order

Study is sequential. A long course can span weeks; its dependents begin only after it finishes. This is a constraint schedule, not a personalised learning-speed estimate.

Weekly allocation

02 / RETRIEVAL, NOT GENERATED ADVICE

Follow each recommendation back to a record.

BM25-style lexical ranking over titles, descriptions and skills. Scores rank text overlap; they are not confidence, source trust or semantic understanding.

03 / IMPLEMENTATION & LIMITS

Useful boundaries, before a larger model.

Implemented here

10 synthetic course records; validated prerequisite DAG; target-skill closure and topological order; split-week hour allocation; self-reported prerequisite warnings; evidence ranking and JSON export. Runs locally in your browser, no saved profile.

Tested locally

72 configurations pass 1,311 assertions covering goal and budget combinations, prerequisite order, hour conservation, unique IDs, missing dependencies, cycle rejection and stable evidence provenance. These are software checks, not learner outcome measurements.

Production proposal

Permissioned LMS adapters, versioned course evidence, access controls, educator approval, multilingual retrieval evaluation and a measured recommendation baseline. No LLM or automatic academic decision should outrun its supporting evidence.