Build an AI-Powered Research Workflow

Turn 10 research papers into a survey outline and research plan using AI tools.

Introduction

30 Second Summary

Starting a research degree means reading stacks of academic papers. Turning that reading into a structured plan your research guide can evaluate is the real challenge.

In this project, you will build a multi-tool research workflow that processes 10 academic papers on agentic AI in IoT edge into two concrete deliverables. You will produce a survey paper outline and a research plan complete with gap analysis and proposed research questions.

What You'll Build

A complete research onboarding pipeline using Zotero, Google NotebookLM, and Claude Projects to transform scattered papers into structured academic output.

By the end of this project, you'll have:

A Zotero library with tagged references and a NotebookLM notebook with structured summaries, a comparison table, and a concept mind map.
A survey paper outline with sections, key arguments, and citation mapping produced through AI-assisted cross-source synthesis.
A research plan identifying literature gaps, framing research questions, and mapping your first 3 months of investigation.
Secret Mission: Use Connected Papers to discover citation neighborhoods around your strongest paper and expand your literature base.

Are there any prerequisites?

You will need a Google AI Pro subscription (which includes NotebookLM with expanded access) and a Claude Pro subscription. You will also need approximately 10 research papers (PDFs or links) on your research topic ready to upload.

Set Up Your Research Toolkit

Your research workflow has three layers, each handling a different job. Zotero manages your references and metadata. NotebookLM synthesizes your sources with grounded citations. Claude Projects reasons across all your papers to build arguments and identify gaps.

In this step, you will install and configure all three tools so they are ready to receive your papers in Step 2.

In this step, get ready to:

Install Zotero 9 with the browser connector for saving references.
Create a NotebookLM notebook with a custom research goal.
Create a Claude Project configured as your research advisor.

Install Zotero 9 and the browser connector

Zotero is the standard reference manager in academic research. It stores your papers, auto-populates metadata (authors, year, DOI), and generates citations in thousands of styles.

Open your browser and go to https://www.zotero.org/download/.
Click the download button for Zotero 9 (Windows 64-bit).
Once the download finishes, run the setup program.
Follow the installer prompts to complete the installation.

What is Zotero?

Zotero is a free, open-source reference manager used by researchers worldwide. It saves papers with full metadata (title, authors, journal, year, DOI) and generates bibliographies in over 9,000 citation styles.

The browser connector lets you save references directly from journal pages (IEEE, arXiv, Springer) with a single click.

On the same download page, find the Zotero Connector section.
Click the connector download link for your primary browser (Chrome, Edge, or Firefox).
Follow the browser prompt to add the extension.

You should now see the Zotero Connector icon in your browser toolbar.

Press the Windows key to open your search bar.
Type Zotero and press Enter to open it.

Zotero opens to an empty library. Next, create a free account so your library syncs to the cloud.

In your browser, go to https://www.zotero.org/user/register/.
Fill in a username, email, and password, then complete the registration.
Back in the Zotero desktop app, click Edit in the menu bar, then click Settings.
Select the Sync tab.
Enter your Zotero username and password in the Data Syncing section, then click Set Up Syncing.

Why sync your Zotero account?

Syncing backs up your library to the Zotero servers and lets you access it from any computer. If your machine crashes, your references are safe in the cloud.

Confirm that the green sync arrow appears in the top-right corner of the Zotero window. This means your account is connected.

Zotero not opening?

Make sure the installer finished without errors. Try restarting your computer and launching Zotero again from the Start menu.

If you are on an institutional computer with security software, you may need to contact IT to allow Zotero to run.

help me troubleshoot why Zotero won't open on my Windows PC

Create your NotebookLM notebook

NotebookLM is Google's source-grounded AI research tool. It reads your uploaded papers and answers questions by citing specific sources, reducing hallucination.

In your browser, go to notebooklm.google.
Sign in with the Google account tied to your Google AI Pro subscription.
Click Create new to start a new notebook.
Name the notebook Agentic AI in IoT Edge - Survey.

Now configure a custom goal so NotebookLM responds with the depth and focus of a research advisor.

In the chat panel, click the configuration icon (gear icon near the chat input).
Paste the following text into the custom goal field:

Act as a PhD-level research advisor specializing in IoT systems, edge computing, and AI agents. Analyze sources for methodological approaches, key contributions, limitations, and research gaps. Always cite which source your claims come from.

What does this custom goal do?

The custom goal tells NotebookLM to adopt a specific persona and response style for every conversation in this notebook. It will focus on methodology, contributions, limitations, and gaps rather than giving surface-level summaries.

The "Always cite which source" instruction forces NotebookLM to ground every claim in a specific uploaded paper, making it easy to trace back to the original text.

Click Save to apply your custom goal.

Your notebook now shows the title and the custom goal is active. You will upload papers here in the next step.

Don't see the configuration icon?

Make sure you are signed in with a Google AI Pro account. The custom goal feature is available to all NotebookLM users, but the interface may look different if you are not signed in.

help me find the custom goal settings in NotebookLM

Create your Claude Project

Claude Projects lets you upload documents and set persistent custom instructions that apply to every conversation within the project. This makes it ideal for deep analytical reasoning across your paper collection.

In your browser, go to claude.ai.
Sign in with your Claude Pro account.
In the left sidebar, click Projects.
Click Create project.
Name the project Agentic AI IoT Edge Research.

Now add custom instructions that tell Claude how to analyze your papers.

In the project settings, find the Custom instructions field.
Paste the following text:

You are a research advisor helping me write a survey paper on agentic AI in IoT edge systems, covering IoT security and 6G aspects. When I upload papers, help me: (1) identify key contributions and methodology, (2) find connections and contradictions between papers, (3) identify gaps in the literature, (4) suggest research questions. Always reference specific papers by author/title when making claims.

What do these instructions do?

Custom instructions persist across every conversation you start inside this project. You do not need to re-explain your research goal each time you chat.

The numbered list gives Claude a structured framework: contributions, connections, gaps, and questions. The final sentence ensures Claude always ties claims to specific papers rather than making unsourced generalizations.

Click Save to apply the instructions.
Confirm that your custom instructions appear in the project settings panel.

Why both NotebookLM and Claude?

NotebookLM excels at source-grounded extraction. It cites the exact source for every claim and offers unique outputs like mind maps and Data Tables.

Claude excels at longer-form analytical reasoning, argument construction, and iterative drafting. Using both gives you grounded facts (NotebookLM) plus deep analysis and narrative structure (Claude).

Don't see the Projects option?

Projects is available on Claude Pro and Team plans. Make sure you are signed in with your Pro account. The option appears in the left sidebar on the claude.ai web interface.

help me find and create a Claude Project on claude.ai

Your research toolkit is ready. You have Zotero for reference management, NotebookLM for source-grounded synthesis, and a Claude Project for deep analysis. Next up, you will import all 10 papers into these tools and start building your literature base.

Ingest and Organize Your Papers

Your research toolkit is ready. Now it's time to feed it the raw material: your 10 papers on agentic AI in IoT edge systems.

Each tool you set up in the previous step serves a different purpose in your workflow. Zotero manages your references and metadata. NotebookLM synthesizes across sources with grounded citations. Claude Projects handles deep analytical reasoning and longer-form drafting. Loading your papers into all three gives each tool the context it needs to help you later.

In this step, get ready to:

Import all 10 papers into Zotero with metadata and tags.
Upload your papers to NotebookLM as sources.
Upload your papers to your Claude Project knowledge base.

Import all 10 papers into Zotero

A reference manager like Zotero does more than store PDFs. It automatically extracts metadata (title, authors, journal, year, DOI) so you can search, sort, and cite your papers properly.

Switch to your Zotero application.
For papers you have as PDFs on your computer: drag and drop each PDF directly into the Zotero library pane (the center column).
Right-click each imported item and select Retrieve Metadata for PDF to auto-populate title, authors, journal, year, and DOI.
For papers you have as links (e.g., IEEE, arXiv, or journal pages): navigate to each paper's page in your browser and click the Zotero Connector icon in your toolbar to save the reference.

What does the Zotero Connector do?

The Connector detects when you're viewing an academic paper page and saves the full reference (title, authors, abstract, DOI) in one click. If the PDF is accessible, it grabs that too.

This is faster than manual drag-and-drop because the metadata comes pre-filled from the publisher's database.

Now organize your papers into a dedicated collection and add tags for filtering.

Right-click My Library in the left panel and select New Collection.
Name the collection IoT Edge AI Survey.
Select all 10 items in your library (Ctrl+A) and drag them into the new collection.
For each paper, right-click the item, select the Tags tab in the right panel, and add at least one relevant tag.
Use these tag categories based on each paper's focus: agentic-AI, IoT-security, 6G, edge-computing.

Multiple tags per paper

Most papers will fit more than one category. A paper on AI agents for IoT security could get both agentic-AI and IoT-security. Adding multiple tags makes filtering more powerful later.

Confirm that your collection shows 10 items, each with a title, authors, and year in the center column.

Metadata not appearing after retrieval?

Some PDFs (especially preprints or scanned documents) lack embedded identifiers. If Retrieve Metadata for PDF doesn't fill in the fields, try searching for the paper by title on Google Scholar, then use the Zotero Connector on the search result page instead.

As a last resort, you can manually fill in the title, authors, and year by clicking on the item and editing the right panel fields directly.

Help me fix missing metadata in Zotero for my PDF imports.

Upload all papers to NotebookLM

NotebookLM needs your papers as sources before it can synthesize across them. It supports PDFs, URLs, Google Drive files, and .docx files natively.

Return to your NotebookLM notebook in your browser.
Click Add sources in the Sources panel on the left.
For PDFs you have locally: upload them directly using the file picker.
For papers you have as links (e.g., arXiv, IEEE, or journal URLs): paste the URL directly as a source.
Wait for NotebookLM to process each source. A checkmark appears next to each source when it is ready.

Why does NotebookLM accept URLs directly?

NotebookLM can ingest web pages as sources alongside uploaded files. This means you don't need to download every paper as a PDF first. If a paper is freely accessible at a URL (like an arXiv preprint), pasting the link is the fastest path.

Confirm all 10 sources appear in the Sources panel, each with a checkmark indicating processing is complete.

Source stuck without a checkmark?

If a URL source fails to process, the page may be behind a paywall or require authentication. Try downloading the PDF from your university or institutional access instead, then upload the local file.

For very large PDFs (100+ pages), processing may take a minute or two. Refresh the page if it seems stuck for longer than that.

Help me troubleshoot a source that won't process in NotebookLM.

Upload papers to your Claude Project

Claude Projects requires file uploads for persistent project knowledge. Unlike NotebookLM, it cannot ingest URLs directly as a source type. You will need the PDF version of each paper.

Return to your Claude Project at claude.ai.
Click Add content in the project knowledge section.
Upload each paper as a PDF.
For papers you only have as links: open each link in your browser, download the PDF (most academic paper pages have a download button), then upload the downloaded PDF to the project.

How does Claude handle all these documents?

Claude Projects uses RAG (retrieval augmented generation) to retrieve relevant sections from your uploaded documents when answering questions. This means it can work effectively across more total content than its 200K token context window might suggest.

Documents are chunked and retrieved as needed, so you don't have to worry about exceeding limits with 10 papers.

Confirm all papers appear in the project knowledge base. You should see each file listed by name.

Upload failing for a PDF?

Some scanned PDFs or very large files may fail to upload. Try re-saving the PDF from your PDF viewer (File > Save As) to create a clean copy. If the file is over 30MB, check if a lighter version is available from the publisher's page.

Help me fix a PDF upload issue in Claude Projects.

All three tools now have your papers loaded and ready. Next up, you will put NotebookLM to work generating structured summaries and comparisons across your entire set of sources.

Generate Summaries and Comparisons in NotebookLM

Your papers are now loaded into all three tools. The next challenge is turning those raw sources into structured knowledge you can actually work with.

NotebookLM excels at source-grounded synthesis. Every response it generates references your uploaded documents directly, which means less hallucination and more trustworthy summaries. In this step, you'll use it to extract structured summaries, build a comparison table, and visualize concept relationships across all 10 papers.

In this step, get ready to:

Generate structured per-paper summaries and save them as a note.
Create a comparison Data Table and export it to Google Sheets.
Generate a mind map showing concept relationships across your papers.

Generate per-paper summaries

A structured summary extracts the same five dimensions from every paper: what problem it solves, how it solves it, its contribution, its limitations, and its proposed future work. This consistency makes comparison easy later.

Switch back to your NotebookLM notebook ("Agentic AI in IoT Edge - Survey").
Click into the chat panel and paste the following prompt:

For each source in this notebook, provide a structured summary covering: (1) Research problem addressed, (2) Methodology/approach, (3) Key contribution, (4) Limitations acknowledged, (5) Future work suggested. Format as a numbered list by source.

What does this prompt do?

The five dimensions (problem, methodology, contribution, limitations, future work) are standard components of an academic paper assessment. By asking for all five in a single prompt, NotebookLM processes each source against the same framework.

The "Format as a numbered list by source" instruction ensures the output is organized paper-by-paper rather than theme-by-theme. This makes it easier to verify accuracy against each original paper.

Wait for NotebookLM to generate the full response. It will reference each source by name.
Save the output as a note by clicking the pin or save icon on the response. This keeps it accessible in the notebook's notes panel for later reference.

You should now see a saved note in your notebook containing a numbered summary for each of your 10 papers. Each entry covers the five dimensions you requested.

Response seems incomplete or cuts off?

If NotebookLM's response does not cover all 10 papers, it may have hit a response length limit. Ask a follow-up: "Continue with the remaining sources." It will pick up where it left off.

If a specific source is missing from the response, check the Sources panel. A source without a checkmark has not finished processing yet. Wait for it to complete, then re-ask the prompt.

NotebookLM isn't generating summaries for all my papers. What should I check?

Create a comparison Data Table

A comparison table lets you see patterns at a glance. Which methodology dominates? Which sub-topics are over-represented? The Data Table feature in NotebookLM synthesizes your sources into a structured table you can export.

In the NotebookLM chat, paste the following prompt to generate a Data Table:

Create a data table comparing all papers across these columns: Paper Title, Year, Primary Focus (agentic AI / IoT security / 6G / edge computing), Methodology Type (simulation / prototype / theoretical / survey), Key Innovation, Evaluation Metrics Used, Identified Limitations

What does this prompt do?

The column choices are intentional. "Primary Focus" reveals how your 10 papers distribute across sub-topics. "Methodology Type" shows whether your literature leans theoretical or empirical. "Evaluation Metrics Used" highlights what the field considers valid evidence.

"Identified Limitations" is the most valuable column for research planning. Papers that share the same limitation are pointing to a gap you can target.

NotebookLM will generate the Data Table as a structured output. Review it to confirm all 10 papers appear as rows.
Export the Data Table to Google Sheets by clicking the export option on the table. This creates a spreadsheet you can share with your research guide.

You should now have a Google Sheets document open with your comparison table. Each row represents one paper, and each column captures a specific dimension of that paper's contribution.

Why export to Google Sheets?

Sheets lets you sort and filter. For example, sorting by "Primary Focus" instantly groups papers by sub-topic. Your guide can also add comments directly in the spreadsheet when reviewing your work.

Don't see the export option?

The Data Table feature and export to Google Sheets are available to Pro and Ultra users. Confirm your Google AI Pro subscription is active by checking your plan at one.google.com.

If the table generates as plain text in chat instead of a structured Data Table, try rephrasing: "Generate a Data Table comparing all papers." The phrase "Data Table" signals NotebookLM to use the structured table feature specifically.

Help me get NotebookLM to generate a Data Table I can export to Sheets.

Generate a mind map of concept relationships

Tables show you facts. A mind map shows you connections. NotebookLM's Mind Map feature visualizes how concepts in your sources relate to each other, revealing clusters of related papers and bridging papers that connect different sub-topics.

In your NotebookLM notebook, open the Studio panel.
Select the Mind Map option to generate an interactive visual diagram of your sources.
Wait for the mind map to render. NotebookLM will process all 10 sources and map concept connections between them.

What is a mind map in NotebookLM?

NotebookLM's Mind Map generates interactive visual diagrams that help you grasp the overall structure of your sources. Each node represents a concept that appears across your papers. Lines between nodes show which concepts are discussed together.

Clusters (groups of tightly connected nodes) reveal papers that share approaches or topics. Bridging nodes (concepts that connect two clusters) point to papers that link different sub-fields. These are often the most interesting papers for a survey.

Click into specific nodes to see summaries of how that concept appears across your sources.
Take note of clusters. Papers that share approaches will group together visually.
Identify bridging papers. These connect different sub-topics (for example, a paper linking edge computing to 6G security) and will be valuable anchors in your survey outline.

Your mind map should display an interconnected diagram with visible clusters. Papers focused on similar topics (like IoT security) will appear near each other, while bridging papers sit between groups.

Mind map looks sparse or disconnected?

If only a few nodes appear, confirm all 10 sources show checkmarks in the Sources panel. Unprocessed sources will not contribute to the mind map.

If the map shows many isolated nodes with few connections, your papers may cover very distinct sub-topics with little overlap. This is actually useful information. It tells you your survey needs to establish connections that the literature has not yet made explicitly.

My mind map doesn't look right. How should I interpret what I see?

You now have three structured views of your literature: per-paper summaries (depth), a comparison table (breadth), and a mind map (connections). Next, you'll use Claude to perform deeper analytical synthesis and build your actual survey paper outline.

Deep Synthesis with Claude

Your NotebookLM notebook now holds structured summaries, a comparison table, and a mind map of how your papers connect. These are excellent reference tools, but they describe each paper individually.

In this step, you'll use Claude Projects to reason across all 10 papers simultaneously. The goal is to move from "what does each paper say" to "what does the field look like, where are the gaps, and what structure should my survey paper take."

In this step, get ready to:

Identify major themes, contradictions, and under-explored areas across your papers.
Generate a structured survey paper outline with papers mapped to sections.
Extract sourced quotes and key findings slotted into your outline as a citation map.

Identify themes and contradictions

A survey paper does more than list what each paper found. It identifies patterns across the literature and surfaces disagreements or blind spots that individual papers cannot see on their own.

Return to your Agentic AI IoT Edge Research project on claude.ai.
Start a new conversation within the project by clicking Start new chat.
Ask Claude to perform a thematic analysis by pasting this prompt:

I've uploaded 10 papers on agentic AI in IoT edge systems. Please analyze them and identify: (1) The 3-4 major themes or research directions across this set, (2) Any contradictions or disagreements between papers, (3) Methodological patterns (what approaches are most common vs. rare), (4) Which aspects of the problem space are well-covered vs. under-explored.

Why ask for contradictions?

Research gaps often hide in the spaces between disagreements. If two papers take opposite positions on how to handle latency at the edge, that tension is exactly where original research questions live.

Claude's custom instructions (set in the previous step) tell it to reference specific papers by author and title. This keeps the analysis grounded rather than generic.

Wait for Claude to generate its response. You should see a structured analysis that names specific papers under each theme and calls out at least one contradiction or methodological gap.

Claude's response is too generic?

If Claude gives broad themes without naming specific papers, reply with: "Please reference specific paper titles and authors for each point." The project's custom instructions should enforce this, but an explicit nudge helps.

If Claude says it cannot access the papers, confirm they appear in the project knowledge section (not just in a regular conversation). Help me get Claude to reference my uploaded project documents.

Generate a survey paper outline

With themes identified, the next deliverable is a structured outline that organizes your papers into a coherent narrative. This is the skeleton of a survey paper you could present to your research guide.

In the same conversation, ask Claude to propose a survey structure by pasting this prompt:

Based on your analysis, propose a structured outline for a survey paper titled 'Agentic AI in IoT Edge Systems: A Survey.' Include: (1) Suggested sections and subsections, (2) Which papers belong in each section, (3) The narrative arc (how sections build on each other), (4) Key arguments or synthesis points for each section. The survey should also cover IoT security implications and 6G enablers where relevant.

What makes a good survey outline?

A strong survey paper outline has a narrative arc. Sections build on each other rather than simply listing topics in isolation. Look for whether Claude's outline flows logically from foundational concepts to cutting-edge applications to open challenges.

The outline should map specific papers to each section. This ensures every section has evidence backing it and reveals if any section is thin on citations.

Review Claude's proposed outline. Check that each section has at least 2-3 papers mapped to it.
Iterate on the outline by asking Claude to improve it. For example, ask it to expand thin sections, restructure if the flow feels wrong, or add a section you think is missing.

Iteration prompts to try

If a section feels thin: "Section 3 only has one paper. Can you suggest how to expand it or merge it with an adjacent section?"

If the flow feels off: "Restructure the outline so that the security section comes before the 6G section, since security is a prerequisite concern."

Once you are satisfied with the outline, copy the final version and save it as a local document for reference.

Outline is too shallow?

If Claude produces only top-level sections without subsections or paper mappings, follow up with: "Please expand each section into 2-3 subsections, and list which specific papers (by author/title) provide evidence for each subsection."

Help me get a more detailed survey outline from Claude.

Extract direct quotes and key findings for citation

An outline with section headings is a start. What transforms it into a usable writing tool is a citation map: specific evidence from your papers slotted into each section, ready for your drafting phase.

Ask Claude to populate your outline with sourced evidence by pasting this prompt:

For each section of the outline, identify 2-3 specific claims or findings from the papers that would serve as key citations. Quote the relevant passages and note which paper they come from.

What is a citation map?

A citation map pairs each section of your outline with specific quoted evidence from your source papers. When you sit down to write, you already know which paper supports which argument. This eliminates the "I read something about this but can't remember where" problem.

Review Claude's output. Each section of your outline should now have 2-3 quoted passages with paper attributions.
Cross-reference a few quotes against your NotebookLM mind map from earlier. Do the clusters in the mind map align with how Claude grouped papers into sections?

Why cross-reference with NotebookLM?

NotebookLM and Claude use different approaches to analyze your papers. NotebookLM is source-grounded (every claim links back to a specific source passage). Claude excels at longer-form reasoning and constructing arguments across documents.

Comparing their outputs helps you catch blind spots. If Claude groups two papers together but your NotebookLM mind map shows them in different clusters, that tension is worth investigating.

Claude not quoting specific passages?

If Claude paraphrases instead of quoting, follow up with: "Please provide direct quotes (exact wording from the papers) rather than paraphrases. Include the paper title and relevant page or section for each quote."

Claude Projects uses RAG to retrieve relevant sections from your uploaded documents. If quotes still seem imprecise, the source PDFs may have formatting issues. Try re-uploading a cleaner version of the paper in question.

Help me get accurate direct quotes from my Claude Project.

You now have a thematic analysis, a structured survey outline, and a citation map with evidence slotted into each section. These are the building blocks of a real survey paper. Next up, you'll turn these insights into a concrete research plan with gap analysis, research questions, and a first-3-months roadmap.

Formulate Your Research Plan

In the last step, you used Claude to pull themes, contradictions, and a full survey outline from your 10 papers. You now have a structured view of the landscape.

Now it is time to turn that synthesis into action. In this step, you will identify what is missing from the literature, frame original research questions, and draft a concrete plan you can hand to your research guide.

In this step, get ready to:

Frame 3-5 research questions grounded in gaps from your literature analysis.
Draft a one-page research plan document with a first-3-months roadmap.
Create a prioritized reading list and save it in Zotero.

Identify gaps and frame research questions

A strong research proposal starts with a gap. Your earlier analysis revealed which aspects of agentic AI in IoT edge are well-covered and which are under-explored. Now you will ask Claude to turn those gaps into formal research questions with supporting justification.

Return to your Claude Project conversation from the previous step.
Send the following prompt to ask Claude for research questions grounded in the gaps it identified:

Based on the gaps you identified in the literature, propose 3-5 research questions that could form the basis of original research. For each question: (1) Explain why it matters (the gap it fills), (2) Suggest a high-level approach or methodology, (3) Note which existing papers provide the closest foundation to build on.

What does this prompt do?

This prompt asks Claude to move from descriptive synthesis (what the papers say) to prescriptive reasoning (what should be researched next). Each question comes with three layers of justification so your guide can immediately see why the question matters.

Claude will produce 3-5 numbered research questions, each with a gap explanation, a methodology suggestion, and references to specific papers from your set.

Switch back to your NotebookLM notebook and open the mind map you generated earlier.
Compare the gaps Claude identified with the sparse areas in your concept map. Look for clusters with few connecting nodes or isolated concepts that no paper addresses deeply.

Why cross-reference two tools?

NotebookLM's mind map gives you a visual, spatial sense of concept density. Claude's analysis gives you a reasoned argument about gaps. When both tools point to the same under-explored area, you have strong evidence that a real gap exists rather than a quirk of one model's interpretation.

Draft a research plan document

Your research guide needs more than a list of questions. They need to see that you understand the landscape, have a clear direction, and know what your first few months of work will look like. This prompt asks Claude to package everything into a single presentable document.

Return to your Claude Project conversation.
Send this prompt to draft your research plan:

Help me draft a one-page research plan I can present to my guide. Structure it as: (1) Research area and motivation (2-3 sentences), (2) Key findings from literature review (bullet points), (3) Identified gaps, (4) Proposed research questions (ranked by feasibility and impact), (5) First 3 months roadmap: what papers to read deeper, which sub-problem to narrow into, and what preliminary work to attempt (e.g., reproducing a key result, building a small prototype, or conducting a focused sub-survey).

What makes this plan effective?

The five-section structure mirrors what a research guide expects to see at the start of a project. Section 5 (the roadmap) is especially important because it shows initiative. It tells your guide you are not just identifying problems, you have a concrete next step in mind.

Review Claude's output. If any section feels thin or generic, ask Claude to expand it with more specific detail from your papers.
Once you are satisfied, download the research plan using Claude's copy or artifact feature. Save it as a document on your computer.

Iterate before you download

You can ask Claude to restructure sections, add more detail to the roadmap, or re-rank questions by a different criterion (e.g., novelty instead of feasibility). Treat this as a collaborative draft, not a final answer.

You should now have a saved document with all five sections filled in and specific papers referenced throughout.

Create a reading priority list

Not all 10 papers are equally relevant to the research direction you have chosen. A prioritized reading list tells you which papers to re-read deeply and which types of papers you still need to find.

In your Claude Project conversation, send this prompt to generate a prioritized reading list:

Given our research questions, which papers from the set should I re-read most carefully? Rank them by relevance to our proposed research direction. Also, what types of papers am I missing? (e.g., do I need more on a specific sub-topic, a specific methodology, or work from a specific research group?)

Why identify missing paper types?

Your initial 10 papers are a starting set, not a complete review. Claude can spot systematic blind spots. For example, you might have plenty of theoretical papers but no experimental prototypes, or papers from one research group but not from their main critics. This awareness guides your next round of literature discovery.

Claude will produce a ranked list of your papers plus a description of what types of papers are absent from your collection.

Copy the full reading priority list from Claude's response.
Switch back to Zotero.
Right-click your IoT Edge AI Survey collection in the left panel.
Select New Note to create a standalone note within the collection.
Paste the reading priority list into the note. Include both the ranked papers and the missing paper types.

Why save this in Zotero?

Zotero is where you manage your references. Keeping the priority list alongside your papers means it is always visible when you open the collection. As you find and add new papers (especially during the Secret Mission), you can update the note to track progress.

You should now see a note item inside your IoT Edge AI Survey collection containing the ranked list and gap analysis.

Don't see the note in your collection?

Make sure you right-clicked the collection name itself (not an individual paper) before selecting New Note. If the note ended up in your general library instead, drag it into the IoT Edge AI Survey collection.

Help me add a standalone note to a Zotero collection.

Secret mission

Expand Your Literature with Connected Papers

Your 10 papers are a starting point, not a finish line. Use citation graph exploration to discover papers your guide didn't share, then update your deliverables with fresh evidence.

Clean Up Your Resources

Decide whether to keep your resources running, pause them to come back later, or delete them entirely.

This project runs entirely on free tools and existing subscriptions (Google AI Pro and Claude Pro), so there are no ongoing costs to worry about.

Resources you used:

Zotero 9 application (installed on Windows).
Zotero Connector browser extension.
NotebookLM notebook "Agentic AI in IoT Edge - Survey" (cloud notebook in your Google account).
Claude Project "Agentic AI IoT Edge Research" (cloud project in your Anthropic account).
Research Plan document (local file).

Keep everything running

No action needed. Choose this if you are actively continuing your research or plan to build on these deliverables with your guide.

Your Zotero library will keep syncing across devices as you add more papers.
Your NotebookLM notebook stays available for querying as you read more deeply.
Your Claude Project retains all uploaded papers and conversation history for future sessions.
Your Connected Papers free account remains active (5 graphs per month) for future literature exploration.
Your Zotero free account continues syncing your library to the cloud.

Pause - I'll come back to this later

Free up local resources while keeping your cloud data intact for when you return.

Close Zotero. Your library stays synced to your Zotero account and will be there when you reopen it.
Your NotebookLM notebook and Claude Project persist in the cloud with no action needed.
Remove the Zotero Connector extension from your browser if you want a cleaner toolbar. You can reinstall it from the Zotero download page when you return.

Delete - I don't want to use this again

Remove all project resources from your machine and cloud accounts. Only choose this if you are certain you will not return to this research topic.

Uninstall Zotero from Windows. Open Settings, select Apps, then Installed apps. Find Zotero in the list and click Uninstall.
Remove the Zotero Connector from your browser. Open your browser's extensions page, find the Zotero Connector, and click Remove.
Delete your NotebookLM notebook. Go to notebooklm.google, find the "Agentic AI in IoT Edge - Survey" notebook, click the three-dot menu on it, and select Delete.
Delete your Claude Project. Go to claude.ai, open Projects in the left sidebar, find "Agentic AI IoT Edge Research", open its settings, and select Delete project.
Delete your Research Plan document from wherever you saved it on your local machine.

Nice Work!

You just built a complete research onboarding pipeline that transforms scattered papers into structured academic output. Your survey outline and research plan are ready to present to your guide.

Here's what you accomplished:

Organized 10 papers with proper metadata, tags, and structured comparisons in Zotero, then generated AI-assisted cross-source synthesis including per-paper summaries, a Data Table, and a mind map in NotebookLM.
Produced a structured survey paper outline with papers mapped to sections and key citations identified using Claude Projects.
Drafted a research plan with gap analysis, ranked research questions, and a first-3-months roadmap covering deeper reading, narrowing your sub-problem, and preliminary work.
Secret Mission: Expanded your literature base using Connected Papers citation graph exploration and updated your deliverables with new findings.

Ready to quiz yourself?