Multi-Agent Literature Review Assistant with Verified Citations for Students
Students writing a literature review often get chatbot answers full of papers that do not exist. Build a multi-agent assistant that plans a search, queries open scholarly databases, reads abstracts and open-access full texts, drafts a structured review, and refuses to cite anything it cannot resolve to a real DOI or arXiv record with a supporting passage. Then measure whether multiple agents actually beat a single agent on quality and cost.
Advanced
Needs a team, real planning, and a supervisor who knows the area.2 semesters · 2 to 3 students
7 to ship
3 optional extrasSuggested stack
What you should ship
- A planner agent that turns a research question into sub-questions and search queries
- Search agents that query OpenAlex, Semantic Scholar and arXiv in parallel and de-duplicate results
- A reader agent that extracts claims, methods and results from abstracts and open-access PDFs, with the passage each came from
- A writer agent that drafts a themed review where every citation links to a verified record and a quoted supporting passage
- A citation verifier that resolves each reference via DOI or arXiv ID and blocks any it cannot resolve
- A single-agent baseline with the same tools, and a comparison of citation accuracy, coverage, cost and time
- An evaluation with at least 20 research questions from your own department, reviewed by two people
If you have time left
- Export to BibTeX and a Word or LaTeX template
- A 'gap finder' view that shows which sub-questions have few papers
- Supervisor mode where a teacher can see which sources a student accepted or rejected
The problem: confident references to papers that do not exist
Final-year and master's students across South Asia start their literature review the same way: ask a chatbot. Studies of ChatGPT-generated bibliographies, such as Walters and Wilder's 2023 paper in Scientific Reports, found that a large share of references were fabricated or contained serious errors. Supervisors now spend time checking whether sources exist at all.
Research tools have improved, but students rarely know which to trust, and many are paid or limited.
What already exists and the gap
- Elicit, Consensus, Scite and SciSpace search papers and summarise findings.
- Deep research features from major AI providers browse the web and write long reports with citations.
- Stanford's STORM (Shao and colleagues, NAACL 2024) generates Wikipedia-style articles with citations using multiple simulated perspectives.
- Anthropic's write-up on its multi-agent research system explains how an orchestrator and parallel sub-agents explore a question.
The gap is a transparent, open source assistant for students built on free scholarly APIs, where citation verification is a hard rule, not a best effort, and where the single-agent versus multi-agent question is measured rather than assumed.
Inspiration
Anthropic's article reports that multi-agent research uses far more tokens than a single chat. That trade-off is the research question in your project: when is the extra cost worth it for a student's review?
What you build
- Planner, search, reader and writer agents with a clear orchestration graph.
- A citation verifier that every reference must pass.
- A web app where a student enters a question and reviews the draft and sources.
The technically interesting part
- Orchestration. Parallel search agents, shared state, and stopping rules so the system does not loop forever.
- Grounded writing. Each sentence that cites a paper must carry a supporting passage. Missing passage means no citation.
- Fair comparison. Same tools, same budget, single agent versus multiple. Measure quality, cost and time.
How you will evaluate it
- Citation validity: share of references that resolve and support the sentence.
- Coverage: overlap with a hand-built list of key papers for each question.
- Cost and time per review, single versus multi-agent.
- Ratings from two reviewers, ideally a supervisor and a senior student.
Path to a product
Universities and students need trustworthy research help that does not invent sources. A free tier for students with paid institutional features (supervisor views, library integration) is a realistic model.
Scope warning
Use only open APIs and open-access texts and respect their rate limits and terms. Do not download paywalled papers. Frame the tool as a starting point for reading, not a replacement for it.
Learn what you need on Sefism
- AI agents roadmap: orchestration, sub-agents and evaluation.
- How we built our multi-agent research system: the design you can learn from.
- AI agent orchestration patterns: choosing between single and multi-agent designs.
- LangGraph documentation: building the agent graph.
- pgvector: storing and searching paper embeddings.
- Should I learn an agent framework or build agents from scratch?: helps you pick your stack.
Get help with this idea
The risk is an expensive system that is no better than one good agent. Tauseef can help you design the orchestration, the citation rules and a fair comparison. If you are new here, create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.