Com6101 Research Agent: A Workflow Automation Agent for Academic Research

Com6101 Research Agent is an educational open-source project that demonstrates how to build a Python agent integrating paper retrieval, automatic summarization, and conversational memory functions to assist in academic research workflows. Originating from the academic course Com6101, this project does not aim to build a production-grade commercial product; instead, it serves as a teaching example to provide developers and researchers with reference ideas for building similar tools.

In academic research, literature review is a fundamental step, but the traditional process is inefficient: it requires switching between multiple databases for searches, filtering large numbers of papers, manually extracting information, and organizing notes. Statistics show that researchers spend an average of over 40% of their working time on literature retrieval and reading, and the information explosion has exacerbated this burden. As an educational project, Com6101 Research Agent aims to demonstrate how to use AI technology to optimize this process.

This agent adopts a modular architecture, with core functions including:

1. Paper Retrieval Module: Supports integration with multiple data sources such as arXiv and Google Scholar, automatically expands query terms (e.g., extending "transformer architecture" to "attention mechanism"), and sorts results based on citation count, publication time, etc.
2. Automatic Summarization Module: Generates hierarchical summaries in the form of one-sentence, paragraph, and structured (problem/method/experiment/conclusion) formats, extracts key information such as research questions and methodologies, and scores the credibility of summaries.
3. Conversational Memory Module: Maintains conversation context (understands anaphora), integrates user notes, and improves retrieval relevance as the conversation deepens.

The technology stack uses Python (due to its AI ecosystem advantages), with core components including LangChain/LlamaIndex (workflow and RAG), OpenAI API/local models (LLM backend), vector databases (semantic search), and SQLite/PostgreSQL (data persistence). Memory management uses a layered architecture: short-term (current conversation context), working (current research session information), and long-term (cross-session knowledge), with information flow implemented via triggers. The RAG architecture process: query conversion to embeddings → vector database retrieval of similar fragments → injection into LLM prompts → generation of evidence-based answers.

As an educational project, its value is reflected in:

• Agent Design: Demonstrates the ability of goal-driven agents to decompose tasks, use tools, and adjust behaviors.
• NLP Applications: Covers core tasks such as text classification, information extraction, text generation, and semantic search.
• Software Engineering Practices: Good practices like modular design, separation of configuration and code, error handling, and unit testing.

Applicable Scenarios: Quickly understanding the field overview in the early stage of literature research, interdisciplinary exploration, teaching demonstrations, and as a foundation for prototype development. Limitations: Not optimized for large-scale literature databases (performance degrades for tens of thousands of entries or more), general LLMs lack depth in professional fields, summary quality depends on LLM capabilities, and there are copyright and database terms issues.

Expansion Directions: Multimodal support (chart/code/dataset processing), collaboration features (team sharing), citation network analysis (knowledge graph), personalized recommendations, and writing assistance (initial draft of literature reviews). Tool Comparison:

Feature	Traditional Literature Management	Commercial AI Tools	Com6101 Research Agent
Automated Retrieval	Limited	Good	Good
Automatic Summarization	No	Yes	Yes
Conversational Interaction	No	Partially Supported	Core Feature
Memory Capability	Static Tags	Limited	Multi-layered Memory
Customizability	Low	Low	High
Production Ready	Yes	Yes	No (Educational Project)

Com6101 Research Agent successfully integrates multiple AI technologies. As an educational open-source project, its value lies in providing learning resources and inspiration. It is suitable for students of AI application development, researchers looking to improve literature efficiency, and tool developers as a starting point to help understand the engineering implementation of agent design, RAG systems, etc. In the future, such tools are expected to lower the threshold for knowledge acquisition and accelerate the process of scientific discovery.