This is the part of the product we and our partner institutions spent the most time on. Lyceum has Assessment Mode on by default: when it detects homework- or exam-pattern questions, it slows down, leans into Socratic prompting, and writes the full session into the student's learning record. Faculty can pull up any student's complete transcript with one click during grading. Think of it as a fully transparent TA, not a ghostwriter.

Every concept explanation is anchored to your uploaded course materials — syllabus, slides, problem sets, assigned readings. When Lyceum gives a definition, it cites the section or page in your textbook so students can flip back. When a question goes outside the course materials, Lyceum says so and routes the student to a human TA rather than guessing.

No. Student conversations are never used in cross-tenant model training, and that's written into our DPA. Faculty can export or delete any student's full record on demand. For institutions that choose private deployment, all data can stay inside your VPC. We're audited under SOC 2 Type II and ISO 27001, with FERPA + COPPA controls baked in.

Typical rollout is 2 – 3 weeks. Week 1 covers SSO/SAML configuration and SIS schema mapping. Week 2 ingests your syllabi, lecture decks, and problem sets. Week 3 is faculty onboarding and a small-cohort pilot. A named Customer Success Manager runs the whole thing. Your IT team typically only attends two 30-minute sessions.

We co-developed a "viewpoint-aware" framework with humanities faculty: when a question has multiple legitimate interpretations, Lyceum doesn't pick one. It surfaces the major schools of thought and helps the student build their own argument. Every humanities answer cites primary sources and recommended readings.

Lyceum is multi-region by default — U.S., EU, APAC. Hybrid deployment lets us keep sensitive data in-region while elastically scaling inference. The Institution tier carries a 99.95% uptime SLA with named incident-response windows. We've operated continuously through Black Friday, finals week, and AP test day.

You don't need a GPU cluster. We support a range of footprints — from a single 8-GPU appliance up to multi-node clusters in your campus cloud. The most common pattern is logical isolation, shared inference: your data stays cleanly partitioned and encrypted, while the compute is hosted on our side. Cost ends up close to public cloud.

Department and Faculty plans include a first-semester pilot clause: if satisfaction comes in below 70%, we refund the unused portion. Institution contracts include a 90-day onboarding window during which either side may terminate at no cost. We'd much rather you make the right decision slowly than the wrong one quickly.