How siddhant-5/researchflow works
This is a Streamlit research-paper evaluator: users upload a PDF, receive an LLM-generated structured review and publishability judgment, and—if publishable—a suggested conference. The main stages are PDF loading and chunking, per-chunk evaluation, aggregation and aspect summarization, then optional venue comparison; the UI offers PDF, JSON, and CSV downloads. Prompt-and-agent composition is the internal LLM layer. The README mentions optional database persistence, but the sampled source does not establish its implementation or wiring, so it is not mapped.
Main entry points: System.py Evaluation orchestration · app.py Streamlit app · Agents.py Agent builders · Utility.py Groq model setup
Components 18
User interface
- Streamlit app
app.py - Report downloads
app.py
Paper evaluation
- PDF loading
PDFManager.py - PDF chunking
PDFManager.py - Evaluation orchestration
System.py - Chunk classification
System.py - Structured evaluation
System.py - Score aggregation
System.py - Aspect summaries
System.py - Paper evaluation model
Utility.py
Venue recommendation
- Venue comparison
System.py - Venue themes
ThemesAndContext.py - Venue decision model
Utility.py
LLM composition
- Agent builders
Agents.py - Prompt templates
Prompts.py - Groq model setup
Utility.py
Other components
- Researcher
- Groq LLM
Connections 22
- Researcher → Streamlit app: uploads PDF
- Streamlit app → Evaluation orchestration: starts evaluation
- Streamlit app → Groq model setup: creates model
- Streamlit app → Venue themes: loads venue context
- Streamlit app → Venue comparison: requests recommendation
- Streamlit app → Report downloads: formats results
- Evaluation orchestration → PDF loading: loads PDF
- Evaluation orchestration → PDF chunking: splits pages
- Evaluation orchestration → Chunk classification: analyzes chunks
- Evaluation orchestration → Structured evaluation: converts analyses
- Evaluation orchestration → Score aggregation: aggregates evaluations
- Evaluation orchestration → Aspect summaries: summarizes aspects
- Evaluation orchestration → Paper evaluation model: builds final result
- Chunk classification → Agent builders: invokes classifier
- Structured evaluation → Agent builders: invokes structured output
- Aspect summaries → Agent builders: invokes summaries
- Agent builders → Prompt templates: composes prompts
- Agent builders → Groq LLM: invokes model
- Groq model setup → Groq LLM: configures client
- Venue comparison → Agent builders: invokes venue agents
- Venue comparison → Venue decision model: returns decision
- Researcher → Report downloads: downloads reports