NexusAI

Inspiration

Modern researchers, engineers, and students spend over 70% of their research lifecycle on manual information management: reading papers in isolation, struggling to connect related discoveries across disciplines, and manually drafting synthesis notes. Crucial connections between concepts slip through the cracks. We built NexusAI as a full-stack intelligence platform to turn fragmented research into a connected, queryable knowledge network.

What it does

NexusAI is an AI-powered full-stack research platform that ingests unstructured research (text notes, whitepapers, or live web URLs), extracts semantic entities and relational connections, maps them onto an interactive force-directed knowledge graph, and provides cited conversational Q&A and executive synthesis reports.

Key capabilities include:

• Full-Stack Ingestion Pipeline: Ingest text or live web URLs with server-side web scraping using Cheerio.
• AI Knowledge Extraction: Powered by Google Gemini 2.0 Flash (with intelligent offline NLP fallback), NexusAI extracts entities across 6 categories (Technology, Concept, Organization, Person, Location, Event) and models inter-entity relationships.
• Interactive Force-Directed Knowledge Graph: Hand-crafted custom SVG physics simulation with glowing nodes, categorized color coding, zoom/pan controls, and real-time relationship tracing.
• Cited Research Chat: Natural language dialogue grounded directly in the user's indexed research materials, returning evidence-backed responses with [Source Title] citations.
• Executive Synthesis Reports: Generates structured reports with Executive Summaries, Key Themes, Critical Entities, Cross-Source Insights, Knowledge Gaps, and Strategic Recommendations.
• Persistent Relational Storage: All research sources, entities, graph edges, chat conversations, and synthesis reports are persisted locally in SQLite with WAL mode.
• One-Click Instant Demo: A pre-seeded curated AI dataset covering Transformer Architectures, RAG & Vector Embeddings, and Multi-Agent Orchestration allows instant exploration.

How we built it

NexusAI is engineered as a robust full-stack application:

• Backend: Node.js & Express RESTful API server.
• Database: SQLite via better-sqlite3 configured with Write-Ahead Logging (WAL) and foreign key cascades.
• Web Scraping: Cheerio-powered server-side scraper with boilerplate stripping.
• AI / LLM Integration: Google Gemini API (2.0 Flash / Pro) with automated schema validation and rule-based heuristic fallback.
• Knowledge Graph Visualization: Hand-crafted SVG force-directed layout engine built with vanilla JavaScript—zero external graph library bloat.
• Analytics & UI: Chart.js for topic distributions, Lucide Icons, and Vanilla CSS with glassmorphic cards, glowing accents, and animated ambient orbs.

Challenges we ran into

• Cross-Source Entity Deduplication: Merging multi-source entity mentions (e.g., "Transformer" appearing across different papers) while aggregating connection graphs and updating description depth.
• Handcrafted Force-Directed Physics: Implementing smooth node repulsion, edge attraction, centering forces, and boundary collisions without D3.js to maintain 60 FPS in pure SVG.
• LLM Output Strictness: Ensuring Gemini returned raw, structured JSON without Markdown fences or conversational noise, handled with regex cleaning and JSON repair fallbacks.
• Offline Resilience: Designing an intelligent rule-based NLP fallback engine so that users can test and explore the full platform even before entering an API key or when offline.

Accomplishments that we're proud of

• Built a complete full-stack web application with SQLite database persistence and server-side scraping.
• Engineered a lightweight, zero-dependency SVG knowledge graph visualization from scratch.
• Created a resilient architecture that supports Google Gemini while remaining fully functional via heuristic extraction when offline.
• Designed a polished, professional UI that rivals production SaaS platforms.

What we learned

• Designing low-latency entity extraction pipelines with Google Gemini 2.0 Flash.
• Building custom force-directed graph physics simulation algorithms from first principles.
• Implementing robust Write-Ahead Logging SQLite architectures in Node.js ES Modules.

What's next for NexusAI

• PDF & ArXiv Integration: Direct upload and automatic parsing of academic PDF whitepapers.
• Dense Vector Embeddings: Integrating hybrid sparse/dense vector search for sub-document semantic chunk retrieval.
• Multi-User Collaboration: Shared collaborative knowledge bases for research labs and engineering teams.

Built With

• chart.js
• cheerio
• css3
• express.js
• google-gemini-api
• html5
• javascript
• lucide-icons
• node.js
• rest
• sqlite
• svg