CLARA

Inspiration: We were inspired by the challenge of navigating and extracting meaningful insights from vast amounts of online content. With AI tools getting more powerful, we wanted to create a smart, agent-powered system that could take a query, explore the web, and deliver precise, contextual summaries — saving users hours of research.

What it does: CLARA (Clinical Agent for Research & Analysis) is an AI-driven web query assistant chatbot that:

Accepts a user’s question or topic. Searches and pulls relevant URLs. Uses LangChain agents to filter, analyze, and summarize content from those URLs. Presents results on a clean, styled HTML interface for easy reading and sharing.

How we built it:

Flask (Python): Built the core backend using Flask to handle user queries, orchestrate the workflow, and serve results. LangChain Agents (Python): Integrated LangChain agents for structured search, filtering, and summarization tasks based on user queries. Web Scraping (Tavily and Google Search API): Implemented custom scraping logic using tavily and google api to extract content from webpages using tools like requests, Selenium, and LangChain’s built-in loaders. Frontend (JavaScript + HTML/CSS): Developed an interactive frontend using HTML and CSS, with JavaScript handling user input and dynamic rendering of the summarization results. Custom Filtering Workflow (Python): Designed a pre-processing layer in Python that filters out irrelevant content before feeding it into the summarizer — improving performance and accuracy.

Challenges we ran into:

Chaining agent steps properly for search → filter → summarize was tricky to orchestrate. Scraped web content often had noisy data (ads, menus, irrelevant sections) which required custom cleaning. Handling long content and multiple sources without hitting token limits.

Accomplishments that we're proud of:

Successfully created a dynamic agentic workflow that handles multiple URLs per query. Removed previous hardcoded limitations on URL counts and summary lengths. Achieved high-quality summaries that retain key points while being readable. Built a full-stack prototype that’s both functional and presentable.

What we learned:

The importance of clean input before summarization — garbage in, garbage out. LangChain agents can be powerful but need careful prompt engineering and structuring. Even simple frontends go a long way in showcasing backend intelligence. Iterative testing and user feedback helped fine-tune both performance and output quality.

What's next for CLARA:

Add file upload support for summarizing local documents (PDFs, text files). Improve filtering mechanisms using semantic relevance scores. Add memory for multi-turn querying and follow-up questions.