TeachTune
A K–12 or community-college instructor drops in their syllabus, slide decks, and old quiz banks, and gets a private AI tutor that speaks in their teaching voice and never wanders off-curriculum.
K–12 and community-college instructors tired of generic AI tutors inventing lessons
- Ingest syllabus PDFs, slide decks, quiz banks, and recorded lecture transcripts in one folder
- Student-facing chatbot that answers only using course material with inline citations to the instructor's own slides
- Per-student adaptive quizzes that pull from the instructor's old questions
- Admin view showing which topics students are stuck on so the instructor can re-teach
LoRA Speedrun's sub-10-minute wall-clock fine-tunes on a single L40S make per-instructor adapter refreshes cheap enough to bundle into an edtech subscription.
K-12 AI education market valued at ~$390M in 2024, projected to reach $7.9B by 2033 (38% CAGR), driven by personalized learning demand.AI In K-12 Education Market Size & Share Report, 2025-2033 ↗Best AI Tutoring Software for School: 2026 Comparison ↗
Crowded field: Khanmigo (Khan Academy-backed, free for teachers), MagicSchool (Free/Plus/Enterprise), Amira, MATHia all target the same teacher pain point with strong distribution.Khanmigo pricing: Free for teachers, $4/month for parents & learners ↗MagicSchool Pricing & Plans ↗
Khanmigo is 100% free for teachers and districts get custom pricing; MagicSchool offers a free tier plus Plus/Enterprise. Charging individual instructors per-adapter subscription is a hard sell against free incumbents backed by Khan Academy.Khan Academy Khanmigo Pricing & Plans 2026 ↗MagicSchool Pricing & Plans ↗
K-12 edtech is a durable, regulation-anched market (FERPA/COPPA gatekeeping) with multi-year district contracts; demand for personalized tutoring is structural and not a fad.Academic Tutoring and Customized Learning Market to Grow ↗
LoRA Speedrun verified as a real GitHub project (Qwen2.5-1.5B on single L40S, sub-10-min fine-tunes plausible), but per-instructor 'voice' + curriculum grounding requires RAG, voice-style conditioning, guardrails, and adapter serving infrastructure — doable but non-trivial.GitHub - Saivineeth147/lora-speedrun ↗