About Levin Lab
Dr. Michael Levin, a distinguished researcher at Tufts University, leads Levin Lab, a pioneering center at the forefront of developmental biology, artificial life, and cognitive science. The lab focuses on understanding how cognition and decision-making emerge across living systems, from cellular networks to synthetic organisms, with a special emphasis on developmental bioelectricity.
By exploring how cells and tissues communicate and coordinate to form and heal complex structures, Dr. Levin’s team seeks the fundamental principles of life and intelligence. Their work blends biology, bioengineering, and computer science, advancing basic science while opening new horizons in biomedicine, regeneration, and synthetic life.
What challenge did Levin Lab face?
- Increase accessibility – Make the lab’s deep scientific content approachable for diverse audiences without sacrificing rigor.
- Boost interactivity – Replace static FAQs with a conversational interface that answers nuanced questions in real time.
- Protect accuracy – Deliver responses that faithfully reflect the lab’s research philosophy and provide verifiable citations.
How did Levin Lab solve it with CustomGPT.ai?
Through close collaboration with Alden and Rui Do Rosario, the team created LevinBot, an AI-driven knowledge agent trained on:
- Peer-reviewed papers, slide decks, and long-form talks authored by Levin Lab.
- A curated set of lab principles that guide answer tone and depth.
Key implementation elements:
- GPT-4 backbone – Advanced language understanding ensures precise, context-aware replies.
- CustomGPT UX customization – Widget styling, color palette, and typography mirror the existing Levin Lab site for a seamless user experience.
- Anti-hallucination guardrails – Source citations appear alongside answers so users can trace claims directly to underlying research, reinforcing trust.
What results did Levin Lab achieve?
- 24 / 7 global availability – Visitors from any region can engage with Levin Lab content at any hour.
- 90 + language support – Automatic translation broadens reach to international students and researchers without extra effort from the lab.
- Seconds-level response time – Users obtain well-sourced answers instantly instead of manually combing through lengthy PDFs or recordings.
- Academic showcase – LevinBot features regularly in live demos and conference talks, illustrating how AI can elevate scientific communication.
“Omg finally, I can retire! A high-school student made this chat-bot trained on our papers and presentations”
| Objective | Before LevinBot | After LevinBot |
|---|---|---|
| Research accessibility | Dense PDFs, keyword search only | Conversational Q&A, instant answers |
| Response time | Hours or days (email, manual search) | Seconds |
| Language access | English-primary content | 90+ languages supported |
| Staff time on repetitive questions | Regular manual effort from Dr. Levin’s team | Automated by LevinBot |
| Availability | Business hours, scheduled sessions | 24/7, no downtime |
Why it Worked
CustomGPT combined a high-performance language model with fine-grained UX control and transparent source citation. This blend preserved scholarly integrity while turning dense scientific material into a friendly conversation. The anti-hallucination safeguards ensure credibility, and the visual consistency helped users trust that LevinBot was an official extension of the lab.
What can LevinBot answer?
| Use case | Example question | User | Value |
|---|---|---|---|
| Research paper discovery | “What papers has Levin Labs published on bioelectric memory?” | Graduate student | Saves hours of manual search |
| Concept explanation | “What is developmental bioelectricity in simple terms?” | Science journalist | Accurate, citable explanation |
| Cross-study synthesis | “How does the xenobot research connect to the planaria work?” | Adjacent-field researcher | Bridges across the corpus |
| Student learning | “What should I read first to understand the lab’s approach to cognition?” | Undergraduate | Curated entry point to the literature |
| Public education | “Why does this lab study worm memory to understand cancer?” | Curious public visitor | Accessible, engaging answer |
| Research communications | “What are the key findings from the lab in the past five years?” | Science communicator | Synthesized, sourced summary |
Conclusion
LevinBot has revolutionized Levin Lab’s digital presence by transforming static academic resources into an interactive, trustworthy, and globally accessible tool. Students, educators, and science enthusiasts can now explore sophisticated biological concepts without deep domain expertise, strengthening the lab’s mission to democratize knowledge.
Want to try out LevinBot for yourself? Visit drmichaellevin.org and try sending your own query!
Frequently Asked Questions
What is an AI research assistant?
An AI research assistant is a software tool trained on a specific body of research content, such as academic papers, presentations, and institutional publications, that can answer questions from that content in natural language. Unlike general-purpose chatbots, a dedicated AI research assistant draws only from verified source material and provides citations for its answers, making it suitable for scientific and academic contexts where accuracy and traceability matter.
How can AI help research institutions?
AI can help research institutions by turning static knowledge libraries into interactive, conversational interfaces. Researchers spend less time answering repetitive questions. Students gain faster access to complex material. International audiences can engage in their native language. Public outreach scales without requiring additional staff. A tool like LevinBot demonstrates all of these benefits in a live research context. The CustomGPT.ai education industry page covers the full range of academic applications.
Can AI answer questions from research papers?
Yes. When an AI assistant is trained using retrieval-augmented generation on a specific corpus of research papers, it can answer detailed questions from that content and cite the specific papers and passages that support each answer. The key requirement is that the AI must be grounded in the actual documents rather than generating answers from general knowledge.
How does LevinBot work?
LevinBot is trained on Levin Labs’ library of peer-reviewed papers, slide decks, and recorded talks. When a user submits a question, the system retrieves the most relevant passages from that corpus, generates a response grounded in those passages, and displays citations so the user can verify the answer against the source material. The assistant is available 24/7 and supports questions in more than 90 languages.
What is the best AI chatbot for scientific research?
The best AI chatbot for scientific research is one that is trained exclusively on a research institution’s own verified content, provides source citations for every response, avoids hallucination by refusing to answer beyond what the source documents support, and can be deployed without a software engineering team. CustomGPT.ai meets all of these criteria and has been validated by institutions ranging from university research labs to large membership organizations.
How does CustomGPT.ai reduce hallucinations?
CustomGPT.ai uses a retrieval-augmented generation architecture in which the AI retrieves content from a specific, curated document library before generating any response. Answers are constrained to what the source material contains, and every response includes inline citations. When the content library does not support an answer, the assistant acknowledges the limitation rather than generating a confident but inaccurate response.
Can universities build AI assistants without coding?
Yes. CustomGPT.ai’s no-code platform allows any team to upload documents, configure an assistant, and deploy it publicly without writing a single line of code. LevinBot at Levin Labs was initially built by a high school student, demonstrating that the platform is genuinely accessible to non-technical users.
Can AI cite research sources?
Yes, when built on a retrieval-augmented platform like CustomGPT.ai. Every response LevinBot generates includes citations that identify the specific document and passage supporting the answer. Users can open the cited source to verify the claim, maintaining the transparency standards that scientific communication requires.
What documents can be used to train a research assistant?
A research AI assistant can be trained on peer-reviewed papers in PDF format, slide decks and presentations, recorded talks and transcripts, institutional reports and white papers, and web-based content including lab websites and blog posts. CustomGPT.ai supports all of these formats natively, with no preprocessing or reformatting required.
Is CustomGPT.ai suitable for academic institutions?
Yes. CustomGPT.ai has been deployed by universities, research labs, professional associations, and public-sector organizations. Its citation architecture, anti-hallucination safeguards, multilingual support, and no-code deployment model make it well-suited to academic contexts where accuracy, trust, and accessibility are the primary requirements. The LevinBot deployment at Tufts University is one of multiple academic use cases in the CustomGPT.ai customer library.
