Research Exploration using Theo

Use these workflows to move from an early research interest or a known paper to a focused review, verified citations, and an up-to-date view of the field.

Choose where to start

• A broad field or early idea — Start with Problem Identification.

• A clear research question or topic — Go directly to Literature Review.

• A request for recent arXiv papers — Start with Paper Radar.

• One to three papers you already know — Add them as sources, then run a focused Literature Review.

• An existing review and a question about what is new — Run Paper Radar against the review.

• A completed report whose references you want checked — Run Citation Verification.

You do not need to name the phase but you are welcome to call out a specific phase. Tell Theo what you have, what you want to understand, and whether you need discovery, deep review, or citation verification. 

Workflow 1: From a broad question to a verified literature review

1. Describe the broad interest. Ask Theo to identify focused, tractable research directions and important gaps.

2. Review the candidate directions and choose one. Theo will not select a research direction on your behalf.

3. Ask Theo to run a Literature Review on the selected direction, focusing on established findings, competing methods, limitations, disagreements, and open gaps.

4. Question the saved review. Ask which findings are strongest, which methods disagree, which evidence is weak, and which sources are most important to read.

5. Run Citation Verification to check whether each cited paper is real and whether it supports the claim attached to it.

Try asking: “I am interested in how generative AI affects scientific reasoning. Help me identify a focused research problem and the most important gaps.”

After choosing: “Carry this direction into a Literature Review. Establish what is known, what is contested, and where the direction remains open.”

For verification: “Verify the citations in this Literature Review. Check that each reference is real and that it supports the claim made about it.”

Workflow 2: From a clear question directly to a literature review

1. State the question and desired scope. Include the domain, population or system, relationship or intervention, outcomes, methods to compare, and any date limits that matter.

2. Inspect the report and ask Theo to synthesize it before starting another search.

3. Run Citation Verification. Keep claim-to-citation alignment on when you need to know whether the evidence supports the writing; turn it off only when you need the faster reference-existence check.

Try asking: “Review evidence on whether retrieval-augmented generation improves factual accuracy in clinical question answering. Compare evaluation methods, datasets, reported effects, and limitations.”

Workflow 3: From Paper Radar to a focused review

1. Give Paper Radar a clear research interest.

2. Ask for an abstract-level synthesis if you only want to understand the themes, differences, and most promising papers in the scan.

3. If you want a formal review, ask for a manageable shortlist and select papers by rank or arXiv ID. Do not carry every result forward automatically.

4. Ask Theo to review the selected papers. Theo creates a compact manifest of titles, arXiv IDs, and links so Literature Review can locate and verify them without loading every PDF into the conversation.

5. Run Citation Verification on the resulting focused review.

Try asking: “Find recent arXiv papers on test-time scaling for diffusion models. Then let me choose which papers should receive a deeper review.”

After the scan: “Group the most relevant results by theme and shortlist the papers most useful for a deeper review.”

After selecting: “Compare the selected papers’ research questions, methods, evidence, findings, limitations, disagreements, and open problems.”

Workflow 4: Start from one or a few known papers

1. Provide the paper URLs and ask Theo to add them as sources.

2. Wait until source processing is complete. A newly added paper may take a short time to become readable.

3. Run a focused Literature Review grounded in those papers. Ask Theo to search for additional sources only where needed for verification or context.

4. Ask paper-specific questions, then verify the citations and claim-source alignment in the resulting report.

Try asking: “Add these two arXiv papers as sources. Compare their methods, findings, assumptions, and limitations, and place them in the context of related work.”

Useful paper-specific questions

• Where do these papers genuinely disagree? You can even invoke Council over her to iterate with Multiple Models.

• Are their evaluation settings comparable?

• Which assumptions drive the difference in results?

• What experiment would best distinguish between their explanations?

A complete example journey

1. Discover: “I am interested in AI tools for scientific discovery. Help me identify a focused, tractable research problem.”

2. Choose: “Carry the direction on evaluating AI-generated hypotheses into a Literature Review.”

3. Review: “Establish what is known, compare evaluation methods, identify disagreements, and surface open gaps.”

4. Understand: “What are the strongest conclusions, and which depend on weak or inconsistent evidence?”

5. Verify: “Run Citation Verification. Check that every cited paper exists and supports the claim attached to it.”

6. Refresh: “Run Paper Radar against the verified review’s open gaps and tell me whether recent work changes the picture.”

7. Update: “Use the new papers I selected to produce an updated review, and explain what changed from the previous version.”

What to tell Theo for better results

• What you are trying to understand or decide.

• Whether you want historical coverage, recent papers, or both.

• The domain, population, method, relationship, or outcome that defines relevance.

• Whether you want discovery, abstract-level synthesis, full-text review, or citation verification.

• Any papers or reports that must be used as context.

• The comparison or decision the final report should support.

A request such as “Find recent papers, let me choose which ones matter, review those deeply, and then verify the citations” is enough for Theo to guide the workflow and pause at the appropriate selection points.