AIRA Research Assistant

Quick Verdict

AIRA Research Assistant is a RAG literature application maintained by Hamid Vakilzadeh. Its canonical source is the hamid-vakilzadeh/lit-review-assistant GitHub repository, not the Streamlit demo URL. The app uses Streamlit to help researchers find, organize, and ask questions about academic literature. As of July 21, 2026, the MIT-licensed repository was not archived and had recent activity, but it had only single-digit stars and no formal releases. Evaluate it as a small academic software project, not a mature research platform.

It best fits technically capable individual researchers who can manage Python, model credentials, and source verification. Free code does not make the complete workflow costless: model and embedding APIs can charge, while local deployment uses compute and storage. Authentication on the hosted Streamlit demo is currently unreliable, and public documentation does not establish clear upload, log, backup, deletion, or model-provider privacy boundaries. Do not upload unpublished manuscripts, confidential research, student records, or regulated material. The separate AIRA-SemanticScholar repository is an MCP literature-retrieval server, not the same application. This directory does not recommend AIRA by default.

Best For

• Python and Streamlit developers studying or modifying a compact literature RAG application.
• Researchers willing to self-host, manage model API keys, and verify every citation against the source paper.
• Teams prototyping with public papers rather than producing publication-ready evidence automatically.
• Not suitable for institutions requiring SSO, role permissions, audit trails, a stable SLA, data-residency contracts, or managed collaboration.
• Not ideal for users unwilling to maintain dependencies, budget model usage, troubleshoot PDF extraction, and detect citation hallucinations.

Key Features

• Streamlit research interface: a browser UI for literature discovery, organization, and question answering.
• RAG document Q&A: chunks and indexes imported material, retrieves relevant context, and sends it to the configured model.
• PDF workflows: builds context from research PDFs, subject to errors with scans, tables, formulas, and references.
• Local deployment: clone the repository, install requirements.txt, and run streamlit run About.py.
• Provider-dependent generation: output quality, data destination, and cost depend on the selected model and API configuration.
• Published design research: the authors published a paper describing AIRA’s design and user evaluation; that is not a security audit.
• Separate MCP component: AIRA-SemanticScholar provides Semantic Scholar retrieval through another repository with its own dependencies and boundaries.

Use Cases

Start with a small set of public papers that you are entitled to process. Build a local index, ask a bounded question such as which datasets and methods the papers use, and verify each answer against page-level source evidence. For broader multi-source retrieval and lawful open-access resolution, Paper Search MCP has a clearer connector model. Elicit offers a more finished evidence-table workflow, while Connected Papers is better for visual citation-neighborhood exploration.

A systematic review still needs a documented search strategy, inclusion and exclusion criteria, duplicate screening, retraction checks, full-text reading, and reproducible records. Treat PDF instructions as untrusted content because prompt injection could ask the model to expose an API key, inspect unrelated directories, or change the research question. Verify title, authors, year, DOI, page, and quoted claim for every generated citation.

Pricing

The repository is free under the MIT License. It has no formal GitHub Release, so a serious pilot should pin an audited commit instead of tracking main. Real self-hosting costs include model and embedding APIs, compute, storage, backups, dependency updates, and human review. A local model can replace some provider charges but introduces hardware and operations costs.

The public Streamlit demo has no separate published price, which does not establish permanent availability, privacy, or reliability. Streamlit hosting, application code, and model providers are distinct processors. The AIRA-SemanticScholar MCP must also be audited independently; do not use the core repository’s license or privacy assumptions as a substitute for reviewing that component and its upstream API terms.

Pros

• Clear canonical GitHub repository and simple MIT licensing.
• Compact Streamlit architecture suited to learning and modification.
• Self-hosting gives more control over the application, index, and original files than an anonymous demo.
• A peer-reviewed paper explains design goals and evaluated use cases.
• Separating the Semantic Scholar MCP clarifies architecture and responsibility.

Cons

• Very small adoption, with single-digit stars, few forks, and no formal release.
• Hosted-demo authentication is unreliable, while privacy, retention, backup, and provider details are unclear.
• Free self-hosting still incurs model, embedding, compute, and maintenance costs.
• PDF parsing, retrieval, summaries, and citations can all be wrong.
• Lacks mature team permissions, audit, support, SLA, and administrative controls.
• The Semantic Scholar MCP is a separate project, not one uniformly validated product surface.

Alternatives

FAQ

What is AIRA’s canonical URL?

This entry uses https://github.com/hamid-vakilzadeh/lit-review-assistant. The former Streamlit URL is a hosted demo, not the source repository.

Is AIRA free and open source?

The core Streamlit repository is MIT-licensed and free to use under that license. Model APIs, embeddings, infrastructure, and third-party content can have separate costs and terms.

Can I process private papers in the hosted demo?

That is not advisable. Authentication reliability and the upload, logging, backup, deletion, and model-provider boundaries are not sufficiently clear.

Is AIRA-SemanticScholar the same application?

No. It is a separately maintained Semantic Scholar MCP server. The core AIRA project is a Streamlit RAG application. Install, audit, and assess them independently.

Can AIRA complete a systematic review automatically?

No. It can assist discovery, organization, and summarization, but it cannot replace a complete search protocol, screening, original-source verification, bias assessment, or reproducible records.

Can a small GitHub project be used in production?

Only after an isolated pilot. Pin a commit, audit dependencies and network calls, test PDF and citation accuracy, and accept responsibility for monitoring, backup, incident response, and upgrades.

Bottom Line

AIRA Research Assistant is a clear, readable, MIT-licensed Streamlit literature RAG project rather than a mature SaaS. The canonical repository is useful for learning and controlled self-hosting experiments. The hosted demo has weak authentication and privacy clarity, and the separate Semantic Scholar MCP must not be folded into the core application’s claims. Begin with public or authorized PDFs, pin the code, document model data paths and costs, and return every research claim and citation to human source verification.