# Product

<!-- impeccable:product-schema 1 -->

## Platform

web

## Users

Academic researchers — people actively writing or reviewing papers — who need to
understand a paper's place in the scientific landscape. Their job: find the
hidden connections a plain search misses, spot citation gaps, and check whether
claims hold up against the literature.

## Product Purpose

CitationEdge turns a single uploaded PDF into a deep, multi-angle analysis of
how that paper relates to the scientific field: which works it builds on, which
connections are missing, which claims are supported or suspect. Success means
researchers rely on it to make research faster and more defensible.

## Positioning

Parallel multi-agent research analysis powered by Graph-RAG. Where search tools
return a ranked list of documents, CitationEdge runs nine specialist agents
against the paper at once and materializes the results as a citation knowledge
graph — uncovering hidden connections, citation gaps, and verification results
that a plain search could not surface.

## Operating Context

Researchers upload a PDF and get a job: nine specialist agents analyze it in
parallel, the job runs in a queue with live status, and results are delivered as
an interactive knowledge graph, structured result pages, and an automated PDF
report with optional email delivery. Users authenticate with email (Firebase
Auth) and return to a jobs list of past analyses.

## Capabilities and Constraints

- PDF upload and visual parsing (including scanned/figure-heavy papers).
- Nine specialist agents covering different analysis dimensions, orchestrated
  in parallel via Groq LLMs.
- Citation knowledge graph in Neo4j with an interactive graph visualization.
- Claim verification, citation-gap detection, and synthesis results.
- Automated PDF report generation and email delivery.
- FastAPI backend (Python), React + Vite frontend, MongoDB persistence.
- Deployed as Docker image on Hugging Face; frontend deploys to Vercel.
- Undecided: pricing, team/collaboration features, non-English literature
  support.

## Brand Commitments

- Product name: CitationEdge.
- Keep the current brand: premium "obsidian" aesthetic — pure black
  backgrounds, white accents, glass cards with subtle grain overlay.
- Typography: Satoshi for body, Instrument Serif for display headlines, DM
  Mono / JetBrains Mono for data and code.
- Status colors: green for success, red for errors. Restraint and monochrome
  contrast are the identity; no gradients-based color decoration.

## Evidence on Hand

- Architecture diagrams: CitationEdge_Architecture*.drawio.
- Test papers: test_paper.pdf, test_paper2.pdf, test_v3.pdf.
- Eval output: eval_results.json, agent_accuracy_report.md.
- No real customer testimonials, case studies, or press — future work must not
  fabricate them.

## Product Principles

1. Depth over search — every feature must reveal what a ranked document list
   cannot (hidden connections, gaps, verified claims).
2. Grounded output — claims and verifications must trace back to the papers;
   never fabricate citations or findings.
3. Speed through parallelism — nine specialists at once, results without
   babysitting; the job queue is a feature, not a workaround.
4. The graph is the product — the citation knowledge graph is the primary
   thinking surface, not an optional view.
5. Restraint is the identity — the black/white serif language is the brand;
   preserve it in all future work.

## Accessibility & Inclusion

No product-specific accessibility requirement has been established.