PeerLens

Agent Architecture

Inspiration

New researchers often struggle to identify publishable research ideas, frequently starting projects only to realize later that their work lacks novelty or publication potential. So, we created a tool that helps researchers think systematically about potential ideas based on existing papers, whether extending methodologies, increasing scope, or identifying unexplored areas. Our vision is to simulate the academic peer review process using AI agents to provide objective, comprehensive evaluation of research ideas before researchers invest significant time and effort.

What it does

PeerLens analyzes uploaded research papers using Google ADK and Gemini API to automatically identify research gaps and extension opportunities. When users select a potential idea, two parallel AI agents engage in a structured debate: a researcher agent (thinking from an implementation perspective) and a peer reviewer agent (evaluating novelty and contribution from a publication standpoint). A synthesizer agent then combines their perspectives to generate a publishability score with detailed justification, helping researchers make informed decisions about which directions to pursue.

How we built it

We leveraged Gemini 2.0 Flash for idea generation, We leveraged Google's Agent Development Kit (ADK) with its core components for parallel evaluation: ParallelAgent - enables simultaneous evaluations by researcher and reviewer agents SequentialAgent - ensures proper pipeline execution (parallel analysis → synthesis) BaseAgent - provides foundation for async implementation and error handling State management - maintains data consistency across all agents

Research Papers → Gap Analysis → Parallel Evaluation → Synthesis Agent → Score ↙ ↘ Researcher Agent Reviewer Agent

Challenges we ran into

Our biggest challenge was implementing OAuth authentication for educational institutions. We wanted to restrict access to users with .edu email addresses using post-login triggers, but given the limited hackathon timeframe, we could only implement basic authentication through Gmail and Microsoft accounts.

Accomplishments that we're proud of

We successfully built a system that generates genuinely sensible research ideas—a remarkable achievement given how complex and nuanced academic research evaluation can be. Our parallel agent architecture with distinct researcher and reviewer perspectives creates meaningful debates about novelty, contribution, and feasibility. The system demonstrates a sophisticated understanding of academic standards and provides actionable insights that could genuinely help researchers avoid dead-end projects.

What we learned

We learned to work with Google ADK and workflow agent concepts, particularly understanding how parallel processing can improve both efficiency and evaluation quality. The parallel architecture provides significant time efficiency and objective evaluation, running researcher and peer reviewer agents simultaneously reduces processing time while ensuring independent perspectives, resulting in more comprehensive and unbiased research idea evaluations with better API resource utilization.

What's next for PeerLens

Our vision is to evolve PeerLens into a comprehensive research ecosystem with collaborative features, grant alignment tools, and AI-powered research assistance that democratizes high-quality research evaluation for researchers