ResearchAI

Research AI is an autonomous agent that transforms high-level thoughts into concrete ideas and actionable workflows and it synthesizes insights and outputs product roadmaps and proposals for timeline.

Comment

Inspiration

We were inspired by a simple frustration: turning ideas into execution is still incredibly hard.

Whether you're:

a student trying to find a research topic a founder brainstorming a product or a team discussing ideas in Slack

…the gap between thinking → structured plan → execution is massive.

Most AI tools stop at:

answering questions summarizing content

They don’t act, synthesize across sources, or adapt to who you are.

We wanted to build an agent that:

feels like a real collaborator — one that understands context, adapts to your role, and produces outputs you can actually use.

What it does

ResearchAI is a dual-mode autonomous agent that transforms high-level ideas into structured, actionable outputs.

Research Mode Analyzes papers to identify gaps & limitations Suggests novel research topics Summarizes recent literature Generates: Research proposals Experiment plans Timelines Product Dev Mode Adapts to user role (PM, SWE, Marketing, Researcher) Generates: Product directions Market analysis (with trends) MVP + full roadmap Timeline Tech stack

Key Differentiator

Instead of long reports, ResearchAI outputs:

presentation-ready slides that are tailored to your role and ready for execution.

How we built it

We combined multiple tools into a modular agent architecture:

Auth0 Handles authentication (Google OAuth) Enables role-aware personalization Airbyte Ingests real-world data from: Slack (team discussions, ideas) Google Drive (papers, docs) Keeps the agent context fresh and continuously updated TrueFoundry Orchestrates agent workflows Handles multi-step reasoning pipelines Enables scalable deployment Vector + Context Layer Stores: Papers Slack messages Docs Enables semantic retrieval across all user data LLM + Prompt System Two-mode system: Research Mode Product Dev Mode Role-aware prompting Structured output generation (slideshow format)

Challenges we ran into

We kept pushing our .env files and we thought we needed to set up a postgres database for Airbyte but there was a way to directly connect using agent engine when we asked the company representatives directly.

Accomplishments that we're proud of

Able to code up all the features we intended to implement using AWS KIRO and using the sponsors integrations.

What we learned

DO NOT PUSH THE .ENV FILE TO THE GITHUB REPO otherwise have to create GEMINI API keys again.

What's next for ResearchAI

Adding more features and more connectors using Airbyte by extending our functionality to take more resources and data into consideration while doing research and making the output a slideshow and not an md file.

Built With

Share this project:

Updates