Personalized Learning Tutor
Outcome
Give each learner a tutor that grounds every explanation, hint and practice item in the district-approved curriculum and the learner's own mastery evidence, adapting pace and representation without ever revealing assessment answers or acting as a grade of record.
| Portfolio attribute | Value |
|---|---|
| Primary persona | Student |
| Primary aggregate | LearnerMasteryProfile |
| Cosmos DB containers | 2 |
| Grounding data | curriculum, content, assessments, learner profile |
| Development estimate | USD 60-220 monthly dev estimate |
| Runtime options | Python FastAPI and C# ASP.NET Core |
Productivity hypothesis
These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.
| Impact type | Measure | Conservative internal assumption | Customer measurement |
|---|---|---|---|
| Productivity | Tutoring task cycle time | 10-18% reduction | (baseline median tutoring-task minutes - assisted median minutes) / baseline median minutes * 100 |
| Quality or risk | Rework rate | 3-8% reduction | (baseline corrected or repeated response rate - assisted rate) / baseline rate * 100 |
Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.
Rationale: Curriculum, assessments, approved content, and learner profiles support faster tailored explanation preparation.
Pilot guardrail: Educators retain control over instructional suitability and all student-impacting decisions.
Customer validation plan
- Baseline the current workflow with representative users and tasks.
- Run the accelerator on the same task mix with human review retained.
- Measure tutoring task cycle time and rework rate with the documented formulas.
- Replace internal assumptions with customer-observed results before a business case.
Why Azure Cosmos DB
This AI solution needs an operational data plane, not only a vector index:
- Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
- Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
- Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
- Change feed reacts to source and operational updates without polling.
- Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.
Domain workflow

| Stage | Actor | Input → output | Approval |
|---|---|---|---|
| 1. Mastery diagnosis from attempt history | Mastery estimator | PracticeAttempt → MasteryState | Policy controlled |
| 2. Standard-aligned content retrieval | Retrieval orchestrator | MasteryState → CurriculumChunk | Policy controlled |
| 3. Scaffolded explanation and hinting | Personalized learning tutor | CurriculumChunk → TutoringTurn | Policy controlled |
| 4. Adaptive practice and misconception tagging | Student with tutor | TutoringTurn → PracticeAttempt | Policy controlled |
| 5. Teacher review of proposed learning path | Teacher of record | LearningPathPlan → approvedLearningPath | Human required |
Technical architecture

Cosmos DB data model
| Container | Hierarchical partition key | Documents | Search paths |
|---|---|---|---|
learnerMasteryProfiles | /tenantId, /learnerId | LearnerProfile, MasteryState, PracticeAttempt, TutoringTurn, LearningPathPlan | Operational fields only |
curriculumLibrary | /tenantId, /standardCode | CurriculumChunk | /chunkVectorEmbedding, /chunkVector |
Query contract
| Query | Container | Scope | Purpose |
|---|---|---|---|
| MasteryGapsForCourse | learnerMasteryProfiles | single-partition | Identify standards below the mastery threshold in a course to drive the next practice set. |
| LevelAppropriateContentRetrieval | curriculumLibrary | single-partition | Retrieve approved explanations for a standard matched to the learner's reading band and preferred representation. |
| MisconceptionTrend | learnerMasteryProfiles | single-partition | Surface repeated misconception tags on a standard so the tutor changes representation instead of repeating the same hint. |
| SessionTranscriptForTeacher | learnerMasteryProfiles | single-partition | Return a cited tutoring transcript for teacher oversight and academic-integrity review. |
Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.
Azure deployment

Clone the source repository, authenticate with Azure, and run:
Set-Location accelerators/education/personalized-learning-tutor
az login
./Deploy.ps1 -Runtime python -Location eastus2
Change python to csharp to deploy the .NET implementation.
Explore the implementation
- Launch signed-in demo
- Implementation article
- Domain data model
- Workflow and API contract
- Python application
- C# application
- Domain container Bicep
- Platform Bicep
- Cost and configuration
- Source repository