Intended for smaller scale projects, these awards aim to support exploration and testing of innovative ideas before a larger project can be formulated. Applications of up to $1000 in OpenAI credits are accepted on a rolling basis, and funding will be awarded throughout the fiscal year (July to June) until available funding is exhausted. The total annual budget is $15,000. Renewals are possible, but will be assessed on a case by case basis.
Application Requirements
In one page please describe your project, its academic merits, goals, and how the use of OpenAI credits will further the project. Please be specific about what services and models from OpenAI you will employ. Please refer to OpenAI’s pricing page as well as their tokenizer when determining how much funding you are requesting.
Eligibility
Individuals who hold principal investigator rights at any Harvard school. (Please note: Applicants from Harvard Medical School must have PI rights in one of HMS’s Quad-based, preclinical departments).
How this works
- After your application is submitted it will go through a technical review by HUIT, if the project is judged to be feasible it will then move on to an academic review.
- The academic review will be assessed by the HDSI for merit as well as alignment with Harvard University Generative AI work.
- Should funding be awarded the HDSI will reach out to establish your access to the OpenAI API.
Funded Research
Beaver: An Academic Research Agent for Zotero
Joscha Legewie, Harvard Faculty of Arts and Sciences
An AI agent embedded in Zotero performs agentic, citation‑backed RAG across a scholar’s library to deliver transparent, verifiable literature assistance.
MAPA‑Enhanced: Democratizing AI‑Powered Pathway Analysis
Peng Gao (Harvard T.H. Chan School of Government)
Improved embeddings streamline multi‑omics pathway analysis, boosting accuracy and cutting redundancy and costs for broad community use.
Human‑like Agentic Systems for Complex Reasoning and Planning
Samuel Gershman (Harvard Faculty of Arts and Sciences)
A planning agent learns symbolic abstractions and code‑level “skills” to generalize across visual environments and solve complex tasks efficiently.
Representing the Consumer in Trade Policy and Politics
Gautam Nair (Harvard John F. Kennedy School of Government)
Analyze 150 years of congressional text, via LLMs, to reveal when and why consumer interests gain representation in trade politics.
SW•Efficiency: Evaluating LLM Capabilities in Code Performance Engineering
Vijay Reddi (John A Paulson School of Engineering and Applies Sciences)
Researchers are testing whether AI can automatically make computer code run faster and more efficiently without breaking it.
Developing an LLM-Assisted Pipeline to Monitor Digital Well-Being Design Features
Kasisomayajula Viswanath (Harvard T.H. Chan School of Public Health)
Building an AI tool to track how social media platforms design features that affect users’ mental health and online well-being
For any questions, please reach out to Kevin Doyle (kevin_doyle@harvard.edu)