Provide personalized adaptive practice, adjusting difficulty in real time
Schools transform how students learn by replacing static worksheets with AI tutors that instantly adapt to each child's skill level. This gives MSPs a high-value, sticky offering that solves teacher burnout and improves student outcomes while driving hardware and SaaS revenue.
The problem today
10+ hours
wasted weekly by teachers manually differentiating assignments
30%
of students falling behind due to one-size-fits-all pacing
Maria Chen is the academic director of a 200-student K–8 charter school in suburban Columbus, Ohio. Her biggest frustration isn't test scores — it's watching her best teachers burn out on differentiation busywork when they should be teaching.
01The Problem
Hand-sorting worksheets by tier compounds across five nights into a full evening of lost recovery time each week.
A broken concept compounds silently through four lessons before Friday's quiz exposes the gap.
Sunday workbook reassignment runs on guesswork, not data, leaving the week's intervention plan unreliable before it starts.
A student who masters material early sits idle long enough that her parents begin pricing alternative schools.
A student who shuts down mid-problem goes undetected until quiz day — 72-plus hours after the failure point.
Without coaching infrastructure, differentiation quality stays invisible until a quarterly test confirms the loss too late to recover.
02The Solution
Solution Brief
Fictional portrayal · illustrative
- Maria runs 12 teachers across 200 students, K–8
- Differentiation done by hand — Sunday nights, paper packets, color-coded tiers
- No system connecting student response data to next-day instruction
- 45 min/night of sorting accelerates teacher burnout
- Struggling students compound confusion across four days undetected
- Gifted students disengage; enrollment risk follows
- Quarterly tests surface problems months after intervention windows close
- Each practice session opens at the student's actual level, not grade level
- Difficulty adjusts in real time — scaffolded hints for struggle, harder problems for mastery
- Monday dashboard flags who needs small-group time before the lesson starts
- Student data, rostering, and LMS integration raise switching costs at renewal
- White-label model repeats across tutoring centers and small districts in the region
“I used to spend my Sunday nights texting teachers about which worksheets to pull for which kids. Now I open a dashboard Monday morning and I already know who struggled over the weekend. My teachers are actually excited about differentiation again because they're not the ones doing all the sorting.”
— Maria Chen is the academic director of a 200-student K–8 charter school in suburban Columbus, Ohio
03What the AI Actually Does
Real-Time Assessment Agent
Tracks every student response during practice and continuously updates a model of what they know and don't know — so difficulty adjusts in the moment, not after a quiz is graded.
Adaptive Tutoring Agent
When a student gets stuck, this agent generates a tailored hint or plain-language explanation based on where specifically the student went wrong — not a generic 'try again' message.
Curriculum Sequencing Agent
Selects the next practice problem from the content library based on what each student is ready to learn right now, keeping every student in their optimal challenge zone without any teacher input.
Mastery Analytics Dashboard
Gives teachers and administrators a live view of each student's progression through learning objectives — surfacing who's stuck, who's ready to advance, and which concepts need whole-class re-teaching.
04Technology Stack
Google Workspace for Education Standard
$0/year (Standard tier is free for qualifying education institutions)
Identity provider (IdP) for student and staff SSO. Google Admin Console manages Chromebook fleet policies, app deployment, and content filtering integ…
Clever Rostering & SSO
$0 for the school (Clever charges application vendors, not schools)
Middleware platform that syncs student rosters, teacher assignments, and class enrollments from the SIS (PowerSchool, Infinite Campus, etc.) to the ad…
GoGuardian Admin + Teacher
$7.50/student/year ($1,500/year for 200 students)
CIPA-compliant web content filtering and classroom management. Filters inappropriate content while whitelisting AI platform domains. Teacher module al…
Google Gemini 2.0 Flash API
$0.10/M input tokens, $0.40/M output tokens; estimated $50–$150/month for 200 students at moderate usage
Primary LLM for high-volume student interactions: generating practice questions, providing immediate feedback, and basic hint generation. Most cost-ef…
OpenAI GPT-4.1 API
$2.00/M input tokens, $8.00/M output tokens; estimated $30–$100/month for complex reasoning tasks
Secondary LLM reserved for complex tutoring scenarios: multi-step problem explanations, Socratic dialogue chains, misconception diagnosis, and content…
Pinecone Vector Database
$0 (free Starter tier for up to 100K vectors) or $50/month (Standard tier for production)
Stores vector embeddings of curriculum content (questions, explanations, worked examples) organized by topic, difficulty level, and prerequisite relat…
LangGraph + LangSmith
$0 for LangGraph; $39/month Developer tier for LangSmith (monitoring, tracing, evaluation)
LangGraph is the agent orchestration framework that manages the state machine for each student session—routing between Assessment, Tutoring, and Curri…
Azure App Service (B2 Plan)
$55/month (B2: 2 vCPUs, 3.5 GB RAM) scaling to $110/month (P1v3) during peak
Hosts the FastAPI backend application running the LangGraph agent orchestrator, student session management, and REST API endpoints consumed by the web…
Azure Database for PostgreSQL (Flexible Server)
$30–$65/month (Burstable B2s: 2 vCPUs, 4 GB RAM, 32 GB storage)
Stores student profiles, knowledge state vectors (BKT parameters per skill per student), session logs, interaction history, teacher configuration sett…
Azure Blob Storage
$5–$15/month
Stores curriculum content assets (images, PDFs, worked example diagrams), student interaction audit logs (FERPA compliance), and exported analytics re…
IXL Learning (Supplementary)
$369/year per 25-student classroom license; ~$5–$10/student/year for district pricing
Optional supplementary adaptive practice platform providing 17,000+ pre-built skill practice items across K-12 subjects. Used alongside the custom AI …
Khanmigo by Khan Academy
Free for teachers; $4/month per learner ($800/month for 200 students, or $9,600/year)
Supplementary AI tutor providing Socratic questioning across Khan Academy's content library. Can be deployed immediately while the custom agent is bei…
05Alternative Approaches
Turnkey Platform Deployment (Khanmigo + IXL)
$9,600/year Khanmigo + ~$3,000/year IXL
Instead of building a custom AI agent, deploy two proven turnkey adaptive learning platforms: Khan Academy's Khanmigo ($4/student/month) for AI-powered Socratic tutoring, and IXL ($369/year per 25-student classroom) for adaptive skill practice with built-in difficulty adjustment. Both integrate with Google Classroom via their respective APIs and support Clever SSO. No custom development required.
Strengths
- Significantly cheaper upfront ($0 development vs. $30,000–$80,000 custom build)
- Deployment takes 2–6 weeks vs. 3–6 months for custom
- Requires L2 technician vs. AI engineer
- Annual recurring cost is comparable to custom cloud costs
Tradeoffs
- Limited customization — cannot use proprietary curriculum
- Cannot modify AI behavior
- Limited analytics vs. fully custom dashboards
- Vendor lock-in risk
Best for: Schools with limited budget that want immediate results, or as a Phase 1 deployment while custom agents are being built in parallel
SchoolAI Managed Spaces
Free tier available; Pro and School plans at custom pricing
Deploy SchoolAI's platform which provides teacher-controlled AI 'Spaces' where educators can create custom AI tutoring experiences with guardrails. SchoolAI handles FERPA/COPPA compliance (SOC 2 certified), offers a freemium model, and provides real-time teacher visibility into student-AI interactions. Teachers create 'Spaces' with custom prompts and curriculum alignment without needing any code.
Strengths
- Free tier available (75 student sessions/day limit)
- No hardware or cloud infrastructure needed beyond existing devices
- Very low complexity — teachers create AI spaces through a web interface, no technical skills required
- MSP role is limited to SSO setup and device management
Tradeoffs
- Less sophisticated adaptive algorithm than custom BKT — relies on LLM intelligence rather than explicit knowledge modeling
- Limited integration with SIS/grade passback
- Good analytics but not as customizable
Best for: Tutoring centers and small schools that want teacher-controlled AI immediately with minimal MSP involvement. Good complement to the custom agent for subjects not yet covered by the custom curriculum
Open-Source Self-Hosted Stack (Moodle + OpenAI API)
$0 software licensing; $50–$150/month cloud hosting; OpenAI API costs similar to custom build
Deploy Moodle as the LMS (free, open-source) with the Moodle AI Subsystem plugin, combined with direct OpenAI API integration for adaptive tutoring. Host everything on a single Azure VM or on-premises server. Use Moodle's built-in quiz engine with adaptive mode supplemented by AI-generated hints.
Strengths
- No software licensing costs ($0 for Moodle)
- Cloud hosting $50–$150/month or use existing on-premises server
- OpenAI API costs similar to custom build
- Excellent LTI support and content authoring tools
Tradeoffs
- Significant MSP labor for Moodle administration and customization
- Moodle has a steep learning curve for administration
- Plugin ecosystem is fragile
- No equivalent to LangGraph's agent orchestration built-in
- Moodle's adaptive quiz mode is basic compared to BKT — uses simple pass/fail thresholds rather than probabilistic knowledge modeling
- AI integration is more bolted-on than native
Best for: Schools already using Moodle, organizations requiring full data sovereignty (on-premises hosting), or international deployments where Clever/Google Classroom are not standard
Microsoft 365 + Azure AI (Copilot-Based)
M365 A3 at $3.25/user/month ($7,800/year for 200 students) or A5 at $8/user/month ($19,200/year)
Leverage Microsoft 365 Education (A3/A5) with Microsoft Copilot for Education as the AI tutor, integrated with Teams for Education as the learning environment. Use Azure AI Services (Azure OpenAI) instead of direct OpenAI API for enterprise-grade compliance and data residency. Build custom adaptive logic using Power Automate flows and Azure Functions rather than LangGraph.
Strengths
- Best-in-class security and compliance (FERPA, COPPA, FedRAMP)
- Azure OpenAI offers data residency guarantees and does not train on customer data by default
- Teams integration provides built-in video tutoring, chat, and assignment management
- Better for MSPs already certified as Microsoft Education Partners
Tradeoffs
- Higher total cost than other options
- Azure OpenAI pricing is identical to direct OpenAI but with added Azure overhead
- Power Automate is less flexible than LangGraph for complex agent logic
- Copilot capabilities are improving rapidly but currently less customizable than direct API access
Best for: Schools already invested in Microsoft 365 ecosystem, districts requiring FedRAMP compliance, MSPs that are Microsoft Gold/Solutions partners looking to maximize CSP margin
Hybrid Approach: Turnkey Platform Now + Custom Agent Later
Turnkey license costs + custom development costs in year one; decreases in year two as custom system replaces turnkey licenses
Deploy IXL or Khanmigo immediately (2–4 weeks) to give students adaptive practice right away, while simultaneously beginning development of the custom LangGraph-based agent system (3–6 months). Once the custom system is validated through pilot testing, gradually migrate students from the turnkey platform to the custom system subject-by-subject. Maintain the turnkey platform as a fallback and for subjects not yet covered by custom content.
Strengths
- Provides immediate value while the custom system is being built
- Custom system can be built without time pressure, resulting in higher quality
- Teachers provide feedback on turnkey platforms that informs custom agent design
- Best of both worlds long-term
- Second-year costs decrease as custom system replaces turnkey licenses
Tradeoffs
- Highest total first-year cost (turnkey licenses + custom development)
- Highest management complexity — running two systems in parallel requires careful change management
- Clear communication with teachers needed about which tool to use for which purpose
Best for: RECOMMENDED approach for most MSP clients. It de-risks the project by ensuring students benefit immediately while the custom system is developed properly. Most education organizations prefer seeing incremental value over waiting months for a big-bang launch
Ready to build this?