Back office ops · Production

ChatLTV: a RAG-powered AI faculty co-pilot for Harvard's MBA entrepreneurship course

The problem

Professor Bussgang had accumulated a large corpus of course material — cases, teaching notes, books, blog posts, and years of Slack Q&A — but no scalable way to make it interactively accessible to students for case preparation. Faculty also lacked time to provide individualized written feedback on final papers.

Workflow diagram · grounded in source
1
Student submits query in Slack
trigger
“allowing each of our 250 students to engage with the chatbot either privately or publicly”
2
RAG retrieves relevant content
ai_action
“The relevant context was retrieved from the corpus, which was stored in a vector database (in our case, we chose Pinecone). The most relevant content chunks were then served to the LLM”
3
LLM generates grounded answer
ai_action
“respond to a student's query by providing an LLM (in our case, we chose OpenAI's ChatGPT v4) with two pieces of information: (a) the question being asked, (b) relevant context that the LLM can use to answer the question”
4
Answer with sources delivered in Slack
output
“the LLM shared the document sources for the answers in the Slack reply to the answer so that students could see the source material references”
5
Faculty reviews student queries before class
human_review
“Each morning before class, I would inspect the admin CMS to see what queries had been made by what students”
6
Custom GPT evaluates final papers
ai_action
“I decided to create a custom GPT called "HBS LTV Feedback", a critical academic evaluator to provide feedback on LTV final course papers and startup ideas”
Reported outcome

Over half the class — roughly 170 students — made over 3000 queries during the semester, with nearly 40% giving a quality score of 4 or 5.
Faculty gained insight into student understanding before each class via the admin CMS. A custom GPT enabled written feedback on around 125 final papers, something that had not been done before due to time constraints.

Reported metrics
Students using chatbotroughly 170
Total queries madeover 3000
Queries per caseroughly 130
Students giving quality score 4 or 5nearly 40%
Show all 10 reported metrics
students using chatbotroughly 170
total queries madeover 3000
queries per caseroughly 130
students giving quality score 4 or 5nearly 40%
students choosing private queries99%
final papers to receive AI feedbackaround 125
setup time for custom GPTtwo hours
lines of code for custom GPTzero
build time for ChatLTVroughly 2-3 person months
corpus sizeroughly 200 documents and 15 million words
Reported stack
ChatLTVChatGPT v4PineconeLangchainAzure OpenAI ServiceHBS LTV FeedbackSlack
Source
https://www.linkedin.com/pulse/ai-professor-harvard-chatltv-jeffrey-bussgang-oiaie/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Over half the class — roughly 170 students — made over 3000 queries during the semester, with nearly 40% giving a quality score of 4 or 5.

What tools did this team use?

ChatLTV, ChatGPT v4, Pinecone, Langchain, Azure OpenAI Service, HBS LTV Feedback, Slack.

What results were reported?

Students using chatbot: roughly 170; Total queries made: over 3000; Queries per case: roughly 130; Students giving quality score 4 or 5: nearly 40% (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Student submits query in Slack → RAG retrieves relevant content → LLM generates grounded answer → Answer with sources delivered in Slack → Faculty reviews student queries before class → Custom GPT evaluates final papers.

Related back office ops cases