✝︎
A System Built on Human Dignity

The AI-Enhanced Classroom

What if every session was analyzed, every student was truly seen, and every teacher walked in prepared? This is the architecture that makes it possible — built from real experience in a real classroom.

6
Actor Roles
9
Core Modules
5
Auto Flows
3
Build Phases
▶ What This System Does
Three capabilities. One seamless loop.
🎤
01 — Capture
Every session, on the record
Teachers record class audio with a single tap. Transcription runs automatically. Student speakers are anonymized and assigned codes. Nothing is lost between the first word and the last.
🧠
02 — Analyze
From transcript to insight
AI evaluates every session against student profiles — engagement trends, participation patterns, concept gaps. Teachers get specific coaching prompts, not just data. The student who went quiet two sessions ago gets flagged.
03 — Act
The right view for every person
Role-gated portals give each stakeholder exactly what they should see. Teachers plan. Students learn. Parents stay informed. Administrators see the pattern across all of it.
Built for Every Stakeholder
Six roles. One system.
Every person in the school community has a defined role — and each sees only what they should see. Privacy is not an afterthought. It’s the architecture.
📚
Teacher
The Primary Power User
Full access to their classes. Real names, AI coaching prompts, per-student profiles, and intervention alerts. Controls what every other role sees.
👤
Student
The Learner
Sees their own profile, project status, and — when the teacher enables it — an AI tutor scoped to course content. Never sees other students.
👪
Parent
The Guardian
Sees only their own child: engagement summary, completed assessments, and teacher-approved AI notes. Nothing more, nothing less.
🏫
Admin
School Leadership
Anonymized aggregate reports — class-level trends, curriculum coverage. Individual records require teacher elevation to access.
📋
Dept. Lead
Curriculum Coordinator
Cross-class analytics within their subject. Anonymized view only — no individual student data without teacher consent.
🛠
IT Admin
System Operations
Manages accounts, permissions, infrastructure, and security. No access to student content. System-layer access only.
Privacy by Design
Built to protect what matters most.
Human dignity isn’t just a value we teach. It’s the architecture we build on.
🔒
Student Anonymization
Every student is assigned a permanent code at enrollment. All records use codes as the primary identifier. Real names live in a separate encrypted table, visible only to teachers and authorized admins.
👁
Teacher as Gatekeeper
No AI output reaches a student or parent without teacher review and explicit approval. The system surfaces insight. The human decides what to do with it. That order never reverses.
📊
Longitudinal, Not Episodic
A single session transcript is noise. The system builds a picture over time — engagement trends, mastery maps, participation history — so teachers see patterns, not snapshots.
“AI should serve as an observational and analytical tool while teachers retain the irreplaceable human role in education.”
✝︎  The Human Code — Course Thesis, St. Theresa Catholic School
Built on Real Experience
Not a theory. Forged in a real classroom.
This architecture was designed through “The Human Code: AI, Ethics & The Image of God” — a first-of-its-kind middle school elective across 24 sessions with real students.

What the course proved

  • Students engage differently with AI feedback than teacher feedback — both matter
  • Pre-class preparation with student data changes in-room teaching in real time
  • Transcript analysis catches the student who is present but silent
  • Parents respond to specificity — a number with context beats a grade alone
  • Teacher judgment is irreplaceable at every decision point, without exception

What we’re building next

  • A recording-to-analysis pipeline completing within 90 minutes of class end
  • AI-assisted lesson planning informed by the last three sessions
  • A per-student intelligence profile accumulating across the full academic year
  • Role-gated portals for teachers, students, parents, and administrators
  • An AI tutor — student-facing — that teachers switch on and off per student
🔒︎
For Architects & Partners
Full Architecture Access
The complete technical specification — all 9 modules, permissions matrix, automated workflows, tech stack, staffing model, and build roadmap.
Contact the course team to request access.
Architecture Access Granted
Full Technical Architecture
Complete system specification — modules, permissions, workflows, stack, staffing, and roadmap.
System Design
Five Design Principles
Every architectural decision flows from one of these five commitments.

🔒 Privacy by Default

Students are never identified by name outside their own portal or their teacher’s view. Every external report uses anonymized codes. Real names live in a separately encrypted table.

🧠 AI as Tool, Teacher as Authority

No AI output reaches a student or parent without teacher review and approval. The system surfaces insight. The human acts on it. This order never reverses.

📡 Capture Everything, Share Selectively

Recordings, transcripts, assessments, and grades all flow in. Teachers control what surfaces to each stakeholder layer and when.

📊 Longitudinal, Not Episodic

One transcript is noise. The value is the trend across a full course — what a student’s engagement looks like over 24 sessions, not just yesterday’s class.

⚡ Push, Don’t Pull

Insights are delivered to teachers proactively — pre-class briefs, post-class summaries, intervention alerts — not buried in a dashboard waiting to be discovered.


Core System
9 Core Modules
The nine functional layers. Each is independently deployable but feeds the others.
Module 01
Identity & Access Control
Manages who can see what. Every data request passes through role validation before it resolves. Student identity table is encrypted separately from all content records. Full audit log captures every access event for compliance.
Google OAuth 2.0Firebase AuthRBAC middlewareRole matrixAnonymization layerAudit log
Module 02
Capture & Transcription Pipeline
Teacher taps record on mobile or tablet. Audio uploads automatically. Whisper API transcribes. AI assigns student codes via speaker diarization. Teacher reviews and corrects attribution before the record is finalized and analysis begins. Target: analysis delivered within 90 minutes of class end.
PWA MediaRecorder APIFirebase StorageOpenAI WhisperSpeaker diarizationTeacher correction UIAuto-trigger on upload
Module 03
AI Analysis Engine
Runs on every approved transcript and assessment. Output: engagement scores per student, concept mastery signals, participation frequency, tone and affect indicators, cross-session trend data, and specific actionable coaching prompts for the teacher about individual students by code.
Claude API (Sonnet)Firestore triggersPer-class analysisLongitudinal trackingCoaching promptsIntervention alerts
Module 04
Lesson Planning Workspace
AI-assisted lesson builder informed by prior session data. Generates a pre-class brief 24 hours before each session: which students to watch, suggested discussion prompts, and differentiation recommendations. Teacher reviews, edits, and approves before it’s finalized.
Claude APIStandards alignmentPre-class briefPer-student differentiation
Module 05
Assessment Hub
Ingests quizzes, worksheets, and tests via typed entry, PDF upload, or photo OCR. AI scores objective items and flags edge cases for teacher review. Teacher finalizes grades. Scores feed back into student profiles and trigger a fresh analysis run.
Google Vision OCRClaude APIPDF ingestionAuto-scoringTeacher overrideGrade sync
Module 06
Student Intelligence Profiles
The longitudinal per-student record. Engagement trend, participation history, concept mastery map, assessment scores, teacher notes, AI insights, and intervention flags. Builds session by session across the full course.
Engagement timelineConcept mastery mapWorking style signalsTeacher notes layerIntervention queue
Module 07
Role-Gated Dashboards
Four separate views — Teacher, Student, Parent, Admin — each with rigorously scoped data access. Teacher dashboard includes AI coaching prompts and intervention queue. Student portal can surface an AI tutor when the teacher enables it.
Next.js App RouterTypeScriptTeacher dashboardStudent portalParent portalAdmin reports
Module 08
Public Websites (Per Grade / Subject)
Static or semi-static sites hosted per grade or subject. Public layer: course description and project showcase with teacher permission. Authenticated layer: course materials, resources, and teacher-approved student work. Consistent design system throughout.
Next.jsNetlify / VercelRole-gated routesProject showcase
Module 09
Notification & Automation Layer
The connective tissue. Triggers post-class analysis, sends weekly engagement reports, fires intervention alerts, routes parent communications, and prompts teachers 24 hours before each session. Runs on serverless functions — no cron jobs, no drift.
Firebase Cloud FunctionsSendGridFCM PushPost-class triggerWeekly digestsIntervention alerts

Access Control
Permissions Matrix
What each role can see and do. Every cell is a deliberate decision.
CapabilityStudentParentTeacherDept. LeadAdmin
View own profileOwn child✓ AllBy code
View other studentsAnon. onlyAnon. only
Real names visibleOwn onlyOwn childBy request
View transcriptsAnon. only
View AI analysisIf enabledIf sharedAggregateAggregate
Use AI tutorIf enabled
Edit lesson plansReview only
Submit / score assessmentsSubmitView onlyAggregate
View grades✓ Own✓ OwnAggregateAggregate
Manage user accounts
Configure AI permissionsPer class
Publish to public sitesWith approval

Full    Conditional    None


Automation
5 Key Workflows
The automated flows that make the system feel seamless rather than manual.
Flow 1 — Post-Class Transcript Pipeline
TriggerStop recording
AutoUpload to storage
AutoWhisper transcription
AutoSpeaker diarization
TeacherReview + correct
AutoAI analysis
PushNotify teacher
Target: full analysis delivered to teacher within 90 minutes of class end.
Flow 2 — Pre-Class Teacher Brief
Trigger24hrs before class
AutoPull last 3 analyses
AutoClaude generates brief
PushEmail + in-app
TeacherReviews + adjusts
Includes: students to watch, engagement trends, suggested prompts, differentiation moves.
Flow 3 — Assessment Ingestion
InputPDF / typed entry
AutoOCR extraction
AutoAI scores items
TeacherReviews + finalizes
AutoGrades → profile
AutoRe-run analysis
Flow 4 — Intervention Alert
AutoDetect drop trigger
AutoClaude generates note
PushAlert to teacher
TeacherDismisses or acts
OptionalParent notification
Triggers: engagement below threshold 2+ sessions, missing 2+ assessments, or zero participation in a session.
Flow 5 — Weekly Stakeholder Digest
TriggerFriday 5pm
AutoCompile week’s data
AutoRole-specific reports
TeacherApproves parent version
PushDeliver to each role

Technology
Tech Stack
All decisions biased toward speed of build, cost of operation, and Catholic school IT capacity.
Auth & Identity
Google OAuth 2.0Firebase AuthCustom RBAC middleware
Database
Firebase FirestoreEncrypted identity tableRealtime listeners
File Storage
Firebase StorageAudio recordingsAssessment PDFs
AI — Core
Claude API (Sonnet)Analysis engineLesson planningStudent tutor
AI — Transcription
OpenAI Whisper APISpeaker diarization
AI — OCR
Google Vision APIAssessment extraction
Web Framework
Next.js (App Router)ReactTypeScript
Hosting
Vercel (main app)Netlify (public sites)Firebase Cloud Functions
Notifications
SendGrid (email)Firebase Cloud MessagingPush API (PWA)
In-Class Recording
PWA (iOS + Android)MediaRecorder APIBackground upload
Analytics
Google Analytics 4Custom event tracking
Security
Firestore security rulesAES-256 (identity table)Audit logging

People & Process
Staffing Model
Five roles. Manageable overhead. The system works for teachers, not the other way around.

📚 Teacher (Course Lead)

Primary operator. Records class, reviews transcripts, approves AI analysis, plans lessons with AI assist, manages student AI access, and approves parent communications.

🛠 IT / System Admin

Sets up Firebase, manages OAuth, provisions accounts, configures Firestore security rules, monitors storage and function costs. High effort at setup; steady state is low-touch.

📋 Curriculum Coordinator

Reviews cross-class analytics, works with teachers on lesson calibration, ensures AI coaching prompts stay aligned with curriculum goals.

🏫 Principal / School Admin

Receives weekly aggregate reports. Can request individual student records (anonymized by default). Approves parent communication policies and AI permission scope.

Teacher Weekly Time Requirements (3-session course)

Pre-class brief review

5–10 min. AI delivers it; teacher reads and adjusts plan if needed.

Transcript correction

10–15 min per class. Speaker attribution review — the highest-touch step.

Assessment review

15–20 min per batch. AI scores objective items; teacher validates edge cases.

AI analysis review

10 min. Scan coaching prompts, dismiss false positives, act on intervention flags.

Parent digest approval

5 min. Review auto-generated parent version, approve or edit before send.

Total: ~45–60 min/week

Decreasing as transcript attribution improves over the course.


Build Plan
Roadmap
Three phases. Twelve months. Build what matters most, first.
1
Months 1–3 — Foundation
Get data flowing. Get teachers using the core loop.
  • Firebase project — auth, Firestore, Storage, security rules, audit log
  • Google OAuth with role provisioning for all four stakeholder types
  • PWA recording app — tap to record, background upload, status indicator
  • Whisper transcription pipeline — auto-trigger on upload, store in Firestore
  • Teacher transcript correction UI — speaker attribution review
  • Claude-powered post-class analysis — engagement, participation, concept signals
  • Teacher dashboard v1 — per-student cards, engagement scores, AI coaching prompts
  • Student anonymization layer — code system enforced across all queries from day one
2
Months 4–6 — Intelligence Layer
Add assessments, planning, and longitudinal tracking.
  • Assessment Hub — typed entry and PDF ingestion, AI scoring, teacher finalization
  • Grade sync — scores feed into student profiles and trigger analysis re-run
  • Student Intelligence Profiles — longitudinal engagement timeline, mastery map
  • Pre-class brief automation — 24hr ahead, AI-generated, teacher-reviewed
  • Intervention alert system — threshold-based triggers, teacher notification queue
  • Lesson Planning Workspace — AI-assisted builder with standards alignment
  • Weekly digest automation — role-specific reports, teacher approval gate for parent version
3
Months 7–12 — Stakeholder Layer
Open the system to students, parents, and admin with appropriate gates.
  • Student portal — own profile, project status, assessment history
  • AI tutor (student-facing) — teacher-enabled toggle per student, scoped to course content
  • Parent portal — own child’s summary, teacher-approved AI notes, grade view
  • Admin dashboard — anonymized aggregate reports, cross-class trend analysis
  • Public websites per grade or subject — role-gated routes, project showcase
  • Dept. Lead cross-class view — anonymized analytics across multiple teachers’ sections
  • Audit log UI — admin can see all data access events for compliance reporting

Explicitly out of scope in v1

  • Live AI transcription — latency and cost make async the right call at this scale
  • Video recording — audio is sufficient; video raises significant FERPA and privacy concerns
  • AI-to-parent direct communication — always through teacher approval gate in v1
  • Multi-school tenancy — solve one school thoroughly before scaling the model
  • SIS / gradebook integrations (PowerSchool, etc.) — Phase 4 or later

Data Architecture
Firestore Data Model
Eight top-level collections. Every flow in the system reads from or writes to one of these.

users

id, role, email, displayName, childCodes[], teacherIds[]

students

code (PK), gradeLevel, teacherId, workingGenius{}, enrolledAt

classes

id, teacherId, date, recordingUrl, transcriptId, status, lessonPlanId

transcripts

id, classId, segments[{speaker, code, text, ts}], approved, corrections[]

analyses

id, classId, studentCode, engagementScore, conceptMastery{}, aiInsights, teacherNotes

assessments

id, classId, type, questions[], studentResponses[], finalizedAt

lessonPlans

id, teacherId, sessionDate, objectives[], aiSuggestions, teacherEdits, status

studentProfiles

code, weeklySnapshots[], longitudinalEngagement[], masteryMap{}, interventionFlags[]


Teacher Perspective
From Inside the Classroom
The methodology is documented. So is what it actually costs to run it well — the prep, the transcripts, the friction, and what the process revealed.
📝
Teacher Reflections — Same Access Code
What It Actually Takes
Three hours for every hour of class. The transcript habit. Two very different course phases. Notes from inside a pioneering AI ethics course.
→ Read the reflections