Building a Smart Research Assistant Agent with Amazon Nova 2 Lite: My Journey in the Amazon Nova AI Hackathon
I built an autonomous Smart Research Assistant Agent for the Amazon Nova AI Hackathon using Amazon Nova 2 Lite on AWS Bedrock. This agentic AI system reasons through complex research queries, intelligently selects and executes tools like PDF analysis and web research, then synthesizes accurate, cited answers—showcasing fast, cost-effective frontier intelligence for real-world knowledge work.
As a developer passionate about making AI truly helpful for knowledge work, I recently participated in the Amazon Nova AI Hackathon (February–March 2026) and built Smart Research Assistant Agent — an autonomous, agentic AI system designed to tackle complex research queries without constant hand-holding.
The project is open-source and available here: https://github.com/Devika24/amazon-nova-hackathon
Why Amazon Nova?
When the hackathon launched, I was immediately drawn to Amazon Nova's family of frontier foundation models, especially the new Amazon Nova 2 Lite. This model stands out for its:
- Lightning-fast performance and low cost — perfect for iterative reasoning loops in agentic workflows
- Strong reasoning capabilities for everyday (and not-so-everyday) tasks
- Seamless multimodal support via Amazon Bedrock, enabling future extensions to images, charts, or documents
- Industry-leading price-performance, which made experimentation affordable during intense hackathon sprints
Nova 2 Lite powers the agent's core intelligence, handling both planning/reasoning steps and the synthesis of final answers.
How the Smart Research Assistant Agent Works
The agent follows a clean 5-step reasoning loop inspired by agentic AI best practices:
- Receive the user's research question
- Reason — Analyze the query and plan next steps using Nova 2 Lite
- Decide — Select the right tool (PDF analysis, web research, data analysis)
- Execute — Run the tool and gather results
- Synthesize — Use Nova 2 Lite again to compile a coherent, well-cited final answer
Key integrations include:
- Amazon Bedrock Converse API to invoke amazon.nova-lite-v1:0 with tuned parameters (maxTokens: 1024, temperature: 0.7 for balanced creativity and accuracy)
- PDF analysis via PyPDF2 — the agent reads real research papers, handles edge cases like blank pages, and answers questions like "Summarize key findings" or "Compare methodologies across sections."
- Simulated tools for web research (e.g., latest trends in agentic AI) and data analysis (e.g., market growth projections) — placeholders ready for real integrations like search APIs
Everything runs in a simple Python CLI interface for easy testing, with a bonus animated demo (demo.html) showing real examples:
- Querying a research paper PDF
- Asking about current agentic AI developments
- Performing mock sentiment/market analysis
Setup is straightforward: clone the repo, install dependencies (boto3, PyPDF2, etc.), configure AWS credentials (the us-east-1 region is recommended for Bedrock), and run python amazon_nova_project.py.
Challenges & Learnings
During the hackathon, I iterated heavily on prompt engineering to make Nova 2 Lite's reasoning reliable across multi-step tasks. Its speed allowed dozens of test loops without breaking the bank, and the consistent output format made tool/action parsing robust.
While the current version uses simulated web/data tools, the architecture is extensible — next steps could include real web search via APIs or even Nova Act for browser automation on research sites.
Why This Matters
Research is time-consuming and fragmented. An agent that autonomously reads documents, cross-references web sources, and delivers synthesized insights could transform how students, researchers, and professionals work. Amazon Nova 2 Lite made this feasible in a short hackathon window, proving how accessible frontier intelligence has become on AWS.
If you're exploring agentic AI, multimodal understanding, or cost-effective reasoning models, check out the repo and try it yourself!
#Amazon-Nova #agentic-ai #awsbedrock #generativeai #hackathon
Thanks to the AWS team and IBM for sponsoring the Amazon Nova AI Hackathon — it was an incredible opportunity to build with cutting-edge models.
What are you building with Amazon Nova? Share in the comments!
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article