SciLens

Helping researchers cut through thousands of papers to uncover clear, actionable insights.

Comment

🚀 SciLens Project Story

🌱 Inspiration

The spark for SciLens came from a painful but eye-opening classroom moment.

Our lecturer once gave us a task: “Research AI for stock prediction and prepare a short report.” We scrambled to gather papers, but quickly hit walls:

  • We couldn’t extract relevant performance evaluations (like CNN vs LSTM accuracy).
  • We didn’t structure findings into clear sections.
  • And worst of all — we forgot to cite sources properly.

Our lecturer wasn’t impressed and threatened to give us a zero.

That moment made us realize the true challenge of research isn’t just finding papers — it’s turning them into structured, explainable insights with reliable citations. That’s when we imagined SciLens:

An AI assistant that transforms overwhelming research into clarity, powered by TiDB’s serverless intelligence.

📖 What it does

SciLensAI is an AI-powered research companion that:

  • Fetches and analyzes research papers.
  • Generates structured reports with Introduction, Key Findings, and Conclusion.
  • Builds comparison tables (e.g., CRISPR vs RNAi).
  • Extracts numbers to create automatic visualizations.
  • Provides a chatbot with RAG (Retrieval-Augmented Generation) that cites real sources.

All of this is backed by TiDB Cloud Serverless, which acts as the memory backbone: storing embeddings, retrieving relevant context, and ensuring scalability.

🛠️ How we built it

  1. Document Ingestion
  • Upload PDFs, Word, or PowerPoint files.
  • Extract and chunk text.
  • Generate embeddings using Gemini models.
  • Store embeddings in TiDB’s vector store.
  1. Hybrid Search
  • Combine semantic vector search + keyword search.
  • Ensures results are accurate and context-rich.
  1. Report Generation
  • AI agent retrieves relevant chunks.
  • Outputs structured Markdown with inline citations and references.
  1. Advanced Features
  • Comparison tables with pros/cons.
  • Automatic plots from extracted numbers.
  1. Frontend
  • Built in Next.js + TailwindCSS with teal/emerald theme.
  • Google OAuth for secure sign-in.

⚡ Challenges we ran into

  • Scaling embeddings and keeping retrieval queries fast in TiDB.
  • Enforcing structured AI outputs instead of freeform hallucinations.
  • Extracting tables and numerical values cleanly from unstructured text.
  • Balancing time between backend innovation and frontend polish.

🏆 Accomplishments that we’re proud of

  • Turning a frustrating classroom failure into a fully working research platform.
  • Seamlessly integrating TiDB Cloud Serverless as both a vector store and a scalable knowledge engine.
  • Delivering features beyond simple search: structured reports, visualizations, tables, and RAG-powered chat.

📚 What we learned

  • How RAG + vector databases fundamentally change the reliability of AI assistants.
  • Why citations and references are non-negotiable in research contexts.
  • That TiDB’s hybrid transactional + analytical engine makes it ideal as both a memory layer and a retrieval engine.
  • The value of building modular AI pipelines: ingestion → organization → generation → explanation.

🔮 What’s next for SciLens

We’re just scratching the surface. Next, we want to:

  • Support image and figure extraction from PDFs.
  • Enable table parsing for more structured data retrieval.
  • Scale up to handle massive paper collections.
  • Add richer embedding and retrieval pipelines for multimodal inputs.
  • Expand visualization features — turning research into charts, graphs, and networks automatically.

And through all of this, TiDB Cloud Serverless remains at the core — giving us scalable storage, fast vector search, and the reliability to keep pushing boundaries.

🌟 Conclusion

SciLens transforms research chaos into structured clarity. With TiDB as its backbone, we built not just a tool — but a workflow for the future of learning and discovery.

What started as a near-zero grade turned into a vision:

Helping students and researchers save time, stay accurate, and unlock deeper insights — with AI + TiDB powering every step.

Built With

Share this project:

Updates