How Langchain Can Democratize LLM Development in Edtech
And live events, the DoE, virtual school, the edtech industry's financial standing, and more!
These days, large language models (LLMs) are becoming household names: outside of the clumsy GPT, we have been introduced to Anthropic’s Claude, Google’s BERT which is used for advanced search algorithms and contextual translations, and Baidu’s ERNIE which provides in-depth content comprehension.
Yes, that’s Bert and Ernie, inspired by the iconic Sesame Street roommates, who were themselves named after characters in Frank Capra’s classic “It’s a Wonderful Life.”
BTW, why do voice agents have futuristic female names like Alexa, Siri, and Cortana, while chat models have male names from the 1940’s? #justasking #equAIity
LLMs have proven to be highly convincing in mimicking human-level conversation, providing constructive feedback, and personalizing learning experiences, and as such, they’ve already been recruited to power a wide variety of edtech tools, including AI-powered tutoring systems, educational material content generation, and personalized learning tools. Just this week, we saw funding announcements for several promising new edtech AI companies like Sizzle.ai, Atypical.ai and Magicschool.ai.
There’s no doubt that LLMs are the engines driving the generative AI revolution — but perhaps it’s time to invent the rest of the car… how can LLMs play nice with all the other technologies they need to succeed? We’ve already covered Code Interpreter and how it unlocks the power of Python. However, Langchain takes LLM application development to the next level, and is dubbed the ‘trendiest web framework of 2023’.
Let’s dive into what Langchain is, what it can do, and how it can be applied to edtech.
What is Langchain?
Langchain is an open-source framework and Javascript library, created by developer Harrison Chase, that allows AI developers to build applications on top of LLMs.
Langchain provides pre-configured connective tissue between LLMs and other tools; you can ‘call’ one of many LLMs with just one line of code, ‘call’ additional tools and components with one more, and string (rather, ‘chain’) them all together into user-facing applications that are more than the sum of their parts.
If developing your first LLM application is like creating your first webpage in HTML, Langchain is like Geocities (or maybe even Myspace, Wix, or Squarespace)- it extends developer capabilities with pre-developed modules without demanding more technical expertise, making it easier for anyone, even non-coding plebes like us, to make powerful Gen AI applications.
Don’t take our word for it — here’s Per Harald Borgen, CEO of Norwegian Edtech Scrimba on Langchain:
“Create an AI agent in 14 lines of JavaScript 🔥. This tiny app takes in a user question, queries Google for up-to-date info, and then uses a calculator to perform math operations if needed. This is all made possible thanks to LangChain's JavaScript library, which is the brainchild of Harrison Chase and by far the easiest way for web devs to get started with LLMs and AI engineering.”
Why is Langchain so useful for building applications with LLMs? Because with its pre-built code, it allows a wide variety of use cases, combining the power of LLMs, APIs and data access.
“LangChain helps developers build LLM applications by abstracting away commonly occurring problems: combining models into higher-level systems, chaining together multiple calls to models, connecting models to tools and data sources, building agents that can operate those tools, and helping avoid vendor lock-in by making it easier to switch language models.” (Source)
Langchain-created LLM applications are:
1. Data-Aware
Langchain apps can instantly reference external sources of data, private or public. Suppose that you want to connect to your own data, from your own document… this could be a book, a PDF file, or even a full database with proprietary information.
Langchain allows you to instantly connect a large language model like GPT4 to your own sources of data. We are not just talking about copying and pasting a snippet from text from your document to a ChatGPT prompt, we are talking about referencing an entire database filled with your own data.
2. Agentic
Langchain applications are not limited to taking an input and providing an answer. They can also take action with the output of that information through the use of external tool APIs.
That means a Langchain application can take actions on its responses: it can send emails, book flights, transfer money… or better yet, help users study and learn. Combined with the ability to access private databases, you can envision an app that references a proprietary syllabus and then creates and sends personalized study material to each learner in a class.
3. Modularized
Langchain apps stand ready to combine pre-built sections of common AI systems (like chatbots, personal assistants, summarizers, translators and more) using simple code.
You may be wondering at this point… ‘who put the ‘chain’ in ‘Langchain’?
For some readers, the term ‘chain’ may create reflexive PTSD from the Web 3.0/Blockchain Revolution (personally for me, it brings up the delicious schadenfreude of seeing Tom Brady and the Winklevoss twins lose their shirts on Crypto).
In the generative AI world, ‘chaining’ refers to stringing together multiple models or tools, often using the output of one model as the input of the next.
Langchain, as an open-source library, offers pre-built chains and a number of other components that allow users to create complex and powerful LLM apps easily:
Wrappers allow users to connect applications to large language models instantly such as GPT4 or open-source models from Cohere or Hugging Face.
Prompt Templates allow users to templatize the prompt input to the LLMs (from langchain import PromptTemplate).
Indexes allow users to extract relevant information for the LLMs by referring to their locations in the vector store.
Chains, as promised, allow users to combine multiple components together to solve a specific task and build an entire LLM application. For example, a chain may take a language model and a prompt template and combine them into an interface that takes input from the user and outputs an answer from the language model.
Sequential Chains can return an output from one model and input it into the next model or tool.
Agents allow the LLM applications to interact with external APIs connected to a wide variety of tools, such as Python (which works in some ways similar to Chat GPT’s Code Interpreter). Agents also allow applications to take actions with external tools instead of just answering questions.
In Summary: Langchain offers a robust framework by which creators can string together generative AI components and external tools into powerful applications that access data. Moreover, Langchain’s ease of use makes LLM application development accessible to a much wider variety of developers.
What Does This Mean for Edtech?
So far, this post has been a couple of non-developers trying our darnedest to explain a semi-technical tool in a way that is as clear and accurate as possible. But the real point here is that Langchain, by lowering the barrier to entry for creating complex LLM-based applications, will have serious accelerating effects on the intersection of generative AI and edtech. Here’s how:
1. Empowering Both Big and Small Edtech Companies
We’ve written previously about how LLMs can even the playing field and allow smaller companies (even ‘micro-companies’) to create powerful tools that compete with mid-sized or incumbent edtech firms.
Langchain opens up new horizons for emerging edtech firms, especially those lacking substantial data resources. By permitting seamless integration of LLMs with open-source data (and more and more robust open data sets are appearing by the day), a small dev team can create heavy-duty applications.
On the other hand, for edtech incumbents, Langchain can offer a rapid way to connect to existing data (such as content libraries, learning data, and student performance metrics) and empowers them to create more intelligent, data-driven solutions.
2. Supercharging Tutoring Systems
Langchain can unlock new possibilities for adaptive learning systems – for example, by allowing a tutoring application to reference an individual student's past interactions and behaviors or learning data sets, adaptive learning systems could offer more structured and personalized content and activities with less overhead.
Langchain's agentic applications could also enable the creation of AI coaches and chatbots that do more than just answer questions. These advanced applications can not only perform complex mathematics, but also suggest the next steps in a student's learning journey, monitor progress, and proactively fill learning gaps.
3. Empowering Individual Educators
Just as web development tools empowered a much wider set of users, including educators, to create web pages, Youtube channels, Teachers Pay Teachers libraries, and Tiktok accounts, Langchain can empower motivated educators to create their own custom LLM applications.
With Langchain, educators have a tool that can truly understand the context of their own specific data sets, rather than just processing general knowledge.
For example, educators could use Langchain to generate personalized assignments that combine information from their own syllabus with LMS databases containing their learners’ individual progress and learning data (privacy issues aside… for now).
4. Uplifting Student Creators
Langchain's user-friendly architecture and easily interchangeable components make it accessible to non-technical users of all ages, encouraging them to experiment with various applications and search for effective solutions. Langchain could democratize the development of AI applications within the educational sector, giving more stakeholders the power to innovate.
Consider a scenario where students, even those without a technical background, could construct their own LLM applications, combining the power of generative AI with a wide variety of coding tools and interfaces.
Right now, a motivated student (or educator) could, for example, build an application that:
Scrapes the text of any Wikipedia page (or uploaded reading)
Converts the text into sample quiz questions
Randomizes elements of those quiz questions (like numbers, word order, phrasing) to build a robust question bank
Autogenerates the answers to all questions using LLMs and calculators
Leverages a chain-of-thought LLM to break those answers down into step-by-step, reasoned explanations
Translates all of this text into multiple languages
Creates a chatbot that the student can ask for help on any given step
Uses an image generator to create corresponding visuals for each element
Uses a video generator to turn the images and text into video walkthroughs
Pushes the quiz system into an easy-to-use user interface with gamified elements
And just like that, teaching assistant achieved… and sharable! Langchain's frameworks and building blocks could spark an era of user-driven application development, where students build edtech tools based on their firsthand experiences.
5. Encouraging Community-Driven Development
By providing students, educators, and edtech developers with accessible and simple frameworks, Langchain could even enable a community-driven development ecosystem. Just as developers can share their code to be reused in any format, educators can share their applications, agents and ‘chains’ to be picked up by others, directly influencing the evolution of Edtech applications.
Upcoming Live Events
Our next Bay Area Edtech Happy Hour is coming up on September 8th in San Francisco! This is a monthly recurring event where we come together as a community to connect, collaborate, and enjoy a drink together! Anyone is welcome to attend.
We’d love to see you there, please RSVP if you plan to attend!
Sponsor Edtech Insiders
As we prepare for the launch of Season 7 of The Edtech Insiders Podcast (more coming soon on the AMAZING line up of guests we have), we are looking for sponsors to support upcoming podcasts, events, and newsletters!
If you or someone you know might be interested, you can explore our standardized sponsorship packages or email us at sponsorships@edtechinsiders.org to learn more. We would love to talk with you about how we can work together!
Top Edtech Headlines
1. AI and Edtech Industry Recovery
On a new episode of The Wall Street Journal’s “Tech News Briefing” podcast, Jennifer Carolan (Co-Founder and Partner at Reach Capital) joined as a guest to discuss how AI in education is fueling growth and funding in the edtech industry.
As edtech has been on the decline in 2023 when it comes to funding, and there is continued speculation about what types of edtech are performing best right now, we love to see an industry leader like Jennifer Carolan highlight the real potential that AI has in the education space not only in terms of educational impact, but in terms of industry wide economic opportunity.
Listen to the full podcast to see what she has to say!
2. Gen Z Values Work Experience Over Degrees
New unprecedented research has come out from YPulse that the majority of young people (Gen Z) don’t think they need a college degree to be successful, and that they believe work experience is more important to their success than a college degree.
Simultaneously, a new LinkedIn study shows that recruiters globally are now five times more likely to search for new hires by skills over higher education.
While traditional college won’t disappear overnight, it is certainly becoming less and less of a standard expectation for Gen Z students who are soon to graduate high school. So, what routes will students that opt out of college choose? Apprenticeships, skills based learning opportunities, or online college options (which have been on the rise even since before the pandemic) are a few of the potential routes.
3. Virtual Learning Controversies in K12
As the K12 school year begins to ramp up again, virtual learning is taking center stage in more ways than one.
One the one hand, as teacher shortages have continued to get worse, some public schools are leaning on virtual teachers to support their classrooms, but not without some serious backlash: parents are outraged that students aren’t being provided teachers in real life, and yet schools are fighting back that having a virtual teacher is better than having no teacher at all.
On the other hand, NYC’s first public virtual high school just launched to offer a virtual option for students who preferred the set up offered during the pandemic. There is some concern that continuous online learning won’t meet the social needs and learning support that is necessary for proper development in K12 settings, however since this model is being offered with intention and not out of reaction like during the pandemic, there is reassurance from faculty that learning outcomes will still be met.
4. Political Threats to the Department of Education and Teachers Unions
Four Republican candidates called for the elimination of the DoE in the first Republican presidential debate of the year, and three candidates made statements threatening teachers unions.
Simultaneously, Republicans in the House have introduced a bill that would freeze hiring at the DoE. While it is not expected to pass, this does add to the growing evidence that there is a strong agenda to defund and/or eliminate the DoE in coming years.
This Newsletter Is Sponsored by Magic Edtech
Magic EdTech has helped the world’s top educational publishers and Edtech companies build learning products and platforms that millions of learners and teachers use every day. Chances are that you're probably using a learning product that they’ve helped design or build!
Companies like Pearson, McGraw Hill, Imagine Learning, American Museum of Natural History have used their help to design or build some of their learning products. Now, Magic wants to bring its pedagogical and engineering expertise to make your key learning products accessible, sticky, and well-adopted.
Recent Edtech Insiders Podcast Episodes
The Edtech Insiders Podcast has been HOT these past few weeks. Make sure you catch up on recent episodes with our amazing guests including:
Marco DeRossi, founder and CEO of WeSchool on how education can be transformed through access to collaborative digital tools.
Dr. Satya Nitta, co-founder and CEO of Merlyn Mind on how specialized AI models are revolutionizing classrooms.
Sam Walder, violinist, computer engineer, and the co-founder and CEO of Trala on leveraging AI to make world-class music education accessible to everyone.
Kristina Ishmael, Deputy Director at the US Office of Educational Technology at the US DOE on AI and national edtech policy.
Funding, Mergers, and Acquisitions
Our latest reporting on funding, mergers, and acquisitions comes from Matt Tower’s publication Edtech Thoughts. Matt does an incredible job of covering the latest funding, news, industry updates, and more! If you love Edtech Insiders, be sure to subscribe to Matt’s newsletter as well.
Funding
Kinjo raises $6.5M / US, Gamified Learning / LiveOak Venture Partners, Silverton, Breyer Capital, Roble Ventures / ETCH Assessment of the deal
Atypical AI raises $4M / US, Personalized Learning / 468 Cap, Accelerator Venture Capital, Ascend VC, Bloomberg Beta, SNR
Creatively Focused raises $3M / US, Special Education / York IE, Mairs & Power Venture Capital, Groove Capital, Gopher Angels
YouMakr.AI raises $500K / UK, Test Prep / Varshney Capital, Warwickshire Investment Group, V3 Investment Syndicate
Pistachio raises €3.25M / Norway, Cybersecurity Training / Signals Venture Capital
Weekday raises $2.2M / US, Talent Network / Y Combinator, Venture Highway
ClassWallet raises $95M / US, K12 SAAS Infrastructure / Guidepost Growth Equity, Education Growth Partners, Lazard Family Office Partners
Zeelo raises $14M / US, K12 SAAS Infrastructure / FlatzHoffman, IREON Ventures
Braid raises $6.8M / US, Gig Economy Infrastructure / A16Z, Initialized Capital
Mable Therapy raises $3.95M / UK, Special Education / Gresham House Ventures
Coachbetter raises €2.7M / Switzerland, Corporate Training / Brighteye Ventures, Zen11Holding
Shaktimaan.ai raises $2M / India, Tutoring / YCombinator, Fundersclub, Goodwater Capital
LanguifyAI raises $250K / Sweden, Classroom SAAS / Norrsken VC
Acquisitions
Noodle acquires Meteor Learning / US, Online Program Management
Archipel Academy acquires Springest / Netherlands, Enterprise LMS
SchoolStatus acquires ClassTag / US, K12 SAAS Infrastructure
Trive Capital and Epic Partners buy out Pryor Learning / US, Corportate Training
Other
Imagine Learning launches $50M venture fund / US, Digital Curriculum
Arco agrees to be taken private / Brazil, K12 School Operator
Edtech Insiders Staff Picks
UM graduate instructors ratify contract, end strike before classes start
‘Knowledge is power’: new app helps US teens read books banned in school
AI Detection Tools Falsely Accuse International Students of Cheating
Thanks for reading Edtech Insiders! Subscribe for free to receive a newsletter in your inbox twice a month and support our work!