-
-
Arch image
Thinkwise AI Agent — Intelligent Idea Prioritization
Tagline
Smartly Prioritizing Innovation — One Idea at a Time
Inspiration
This project was inspired by the challenge many organizations face when trying to make sense of hundreds of ideas submitted via internal portals like ServiceNow. We wanted to build something that not only automates idea evaluation but mimics the reasoning process of a human product manager — consistently and transparently.
What it does
The Thinkwise AI Agent reads idea submissions, evaluates them based on implementation effort and ROI, and outputs the top ideas along with transparent, LLM-driven reasoning. It presents the results in an interactive dashboard with adjustable scoring weights.
How we built it
- Frontend: Built using React for a responsive UI that allows idea uploads, score visualizations, and parameter tuning.
- Backend: FastAPI handles requests and coordinates the LangGraph-powered reasoning agent.
- AI Agent: Implemented using Google Gemini 2.0 Flash through the ReAct framework, integrated with scoring tools.
- Database: MongoDB stores input data and analysis history for continuity and scalability.
Challenges we ran into
- Designing a flexible ReAct agent that could think and act across nested workflows.
- Managing large idea datasets efficiently while preserving performance.
- Developing a fair and adjustable scoring formula for ROI vs effort.
- Building robust input handling for both CSV and JSON formats with incomplete or noisy data.
Accomplishments that we're proud of
- A fully functional ReAct AI Agent capable of multi-step reasoning and tool use.
- A clear, user-friendly interface that reveals not only the outcome but the "why" behind it.
- Successfully visualized and ranked ideas based on meaningful business metrics.
- Developed a modular architecture that's easy to scale or adapt to different domains.
What we learned
- Deep integration of LLMs into real-world evaluation processes.
- Building workflows with LangGraph that support iterative logic and decision trees.
- Creating explainable AI solutions that don't feel like black boxes.
- Importance of good UX in technical tools — especially in AI interfaces.
What's next for Repati kosam
- Introduce a feedback loop that helps the agent learn from user preferences.
- Integrate with live data sources like ServiceNow or Jira.
- Enhance the UI with real-time analytics and storytelling features.
- Optimize score evaluation using machine learning or reinforcement learning techniques.
Log in or sign up for Devpost to join the conversation.