Custom AI, web and mobile builds for schools, universities and edtech products

Four things, mostly. Assistants grounded in an institution's own documents — enrolment enquiry, student services, advising, internal staff lookup. Learning and assessment tooling — tutoring support, rubric-based grading drafts, feedback, question banks — with a human at the point where it counts. Native and web apps for students, parents and staff. And the integration layer underneath: SIS, LMS, rostering, single sign-on and reporting. We are not a student information system of record, we do not sell an LMS product, we supply no devices or classroom hardware, and we do not run proctoring or invigilation as a service.

No — there is no named education engagement to point at, and we will not re-badge an adjacent project as one. What transfers is real and checkable: retrieval and extraction over large document sets where the answer had to cite its source, an AI front desk handling intake, scheduling and verification for a service business, native iOS and Android apps, offline-first mobile work for people who lose signal, and multilingual builds for Indian users. Those are the four capabilities most of this vertical draws on, which is exactly why education is on our list rather than a claim about our record.

Neither of those is a certification anybody issues, and we hold no security, privacy or accessibility certifications of any kind. FERPA obligations sit with the institution; a vendor works under a data privacy agreement and, where it applies, the school-official exception. COPPA obligations attach to the operator of a service directed at children under 13, usually with the school standing in for parental consent. What we do is engineer to those constraints: deciding which record types may reach a model at all, keeping student records out of any training or fine-tuning path, contractual no-retention terms with model providers, role-based access, retention and deletion schedules, region-pinned storage, and audit logs. If your procurement requires an audited certification from the vendor, we do not have one, and you should weigh that before we both spend a month on it.

You can reduce it a great deal and contain what is left; you cannot eliminate it, and any vendor who tells you otherwise is selling. The controls that work are structural: a closed, curated corpus rather than the open web, answers that cite the specific page they came from, explicit refusal when the retrieved material does not support an answer, hard handoff to a human on fees, safeguarding, accommodations and anything financial or disciplinary, and an evaluation set of real questions with graded expected answers that runs before every release. We build the logging and review queue alongside it, because the interesting failures are the ones nobody reported.

That is a design decision before it is a technical one, and we make you take it explicitly. In practice it means shaping the interaction toward hints, worked analogues and next-step prompts rather than final answers, scoping the assistant to the material a course actually covers, letting an instructor set what is off-limits per assignment, and being honest in the interface about what the tool will and will not hand over. It also means acknowledging the limit: a determined student can use a general model in another tab, so the design goal is a tool that is more useful than cheating for the student who wants to learn, not a fence.

Usually, but the honest answer depends on your licence and edition rather than on the vendor's marketing page. Some student information and LMS platforms expose good APIs; some gate the useful endpoints behind a partner programme we are not part of; some give you a nightly file drop and call it an integration. We check what your own account actually exposes before scoping, and we design for the degraded case — scheduled OneRoster or CSV transfer with reconciliation, idempotent syncs, and explicit handling for the mid-year enrolment change that arrives in one system and not the other. Where write-back is not permitted, we say so rather than building a queue that silently fails.

We build to WCAG 2.2 AA as the working target — semantic structure, keyboard paths, focus management, contrast, captions and transcripts for media, and accessible maths where the subject needs it — and we test with a screen reader rather than only with an automated scanner, because scanners catch roughly a third of what matters. On the paperwork: an accessibility conformance report is a statement the product owner publishes about their own product. We can run the testing, document the findings honestly including what does not conform, and draft the report for you to review and sign. We will not sign a conformance claim on your behalf, and for high-stakes public procurement an independent third-party audit is the stronger answer than anything the build team writes about itself.

Be realistic with us about this one. District procurement usually wants a vendor already on a state data privacy agreement registry, with audited certifications and a domestic support footprint — and on paper a small studio in India loses that comparison regardless of the code. Where we fit well is edtech product companies building the thing districts eventually buy, private and international schools, tutoring, coaching and test-prep businesses, and individual university departments, faculties or research groups with their own budget. If a district-wide contract is the requirement, tell us in the first call and we will say whether it is worth either of our time.

Scoped in phases against a defined outcome — a discovery and architecture pass that includes the data and privacy decisions, then a build increment you can put in front of real students or staff, then iteration against what they actually did with it. Pricing is scope-based and quoted per engagement. Support runs business hours IST, Monday to Friday; if you need cover through a results day, an enrolment deadline or a term-start weekend, that is a separate arrangement we plan for rather than something we quietly imply.