When Kavitta Ghai and Jordan Long were struggling through their undergraduate degrees six years ago at the University of California, Santa Barbara, the pair came up with Nectir: a teaching assistant for students feeling lost in their coursework.
“I am autistic and I have ADHD. No classroom that I’ve ever been in before has felt comfortable for me or my brain to be in,” Ghai said. “And then I got to college, and I started paying $40,000 to be really, really uncomfortable.”
Nectir, a Playa Vista-based artificial intelligence company for the education space, announced early December it raised $4 million in seed funding, bringing the startup’s total funding to $6.3 million. The seed round was led by Long Journey Ventures with participation from the likes of Santa Barbara-based Entrada Ventures and Behind Genius Ventures, which is based in Mid-Wilshire.
On its head, Nectir is a student help portal that uses OpenAI’s large language model to answer questions. But the company’s core defensible tech isn’t its Google-for-classroom platform. Schools are subjected to the Family Education Rights and Privacy Act, a set of compliance rules that are meant to protect students’ information. Nectir creates a “shell” that can make any technology FERPA-compliant, allowing schools to use and regulate how students interact with new educational tools while keeping student data safe from any LLM’s repository of training data.
“These are files that only that training facility would see, but (schools) can safely put it into our tool because they know it’s not going to go anywhere else,” Ghai said. “And the actual LLM, ChatGPT and OpenAI, they’re never going to see those files.”
Behind the model
The company’s core product underwent several evolutions: first it was a Slack system for classrooms that allowed students and teachers to ask and answer questions around the clock. Then Nectir created Soma, by which students could ask a question and the platform would respond by referencing a repository of teachers’ syllabi, rubrics, textbooks and assignments.
In October 2023, Stanford University approached Nectir with an ask: the higher education institution wanted infrastructure that could use any large language model on the market to create custom AI assistants for specific courses or departments. Thus began six months of testing Nectir’s newest beta product at several universities – including Los Angeles Pacific University – to pressure test the AI and see if it would break.
Now, Nectir is active at 100 schools and with 80,000 students. Through Nectir, students can ask a question and the platform will provide information based on the teachers’ documents (though, the company said, the LLM may fill in blanks on its own if sufficient information isn’t provided).
“The more effective products will be trained on much smaller data sets and some localized documents,” said Stephen Aguilar, an associate professor of education at the University of Southern California. “If you happen to have a population of teachers who document their lesson plans, there’s a lot of really good stuff in there. And then you train a very large language model on those documents. Then you’re going to be in this closed system that’s a little bit more controlled. Students are more likely to get valid information.”
The big picture
Educational technology, often shortened to edtech, was once heralded as the solution to the mounting problems in classrooms: dwindling school budgets, overloaded classrooms and poor pay, compounded by health care, poverty and financial stress experienced outside the classroom, led to teacher burnout and lower educational outcomes for students.
But study after study has found the broad, amorphous category of “educational technology” to have little or inconclusive impact on students’ ability to retain and understand what they’re learning. And while many of these products are geared to help teachers manage increasingly unwieldy classrooms, many edtech products contribute to educator burnout.
Nonetheless, a study from the Los Angeles Pacific University found that students saw a 20% boost in GPAs an average final score raise of 13% after using Nectir for one term.
“(Edtech is) a combination of what is the software meant to do, who are the people using it, and what sort of infrastructure is in place to enable its use,” Aguilar said. “It sort of is this three part process that has to be mostly working in harmony in order for the software to even hope to be effective.”
