Nexus

Inspiration: The Coherence Crisis

• Startups experiencing hyper-growth frequently double their employee intake, leading to severe knowledge fragmentation.
  
• Mission-critical data becomes siloed and scattered across disparate applications, including Slack, Notion, spreadsheets, and enterprise emails.
  
• Founders and teams are forced to execute strategic decisions based on messy, contradictory data lakes.
  
• The fundamental reality of the modern scaling enterprise is that startups do not lack information—they lack coherence.
  

What It Does

• Nexus is a local-first, privacy-preserving Single Source of Truth engine built specifically for high-growth startups.
  
• Because cap tables, intellectual property, and financial forecasts are highly sensitive, the entire platform is powered by local AI via Ollama.
  
• Zero data exfiltration occurs; no proprietary enterprise data ever leaves the founder's machine.
  
• It autonomously ingests raw, chaotic data and deploys AI agents to categorise it into core business verticals, such as Financials, Compliance, Product, HR, and Legal.
  
• It maps structured data into an interactive, Obsidian-inspired Knowledge Graph, synthesising dynamic relationships between previously isolated nodes.
  
• The system proactively audits the database to flag critical business contradictions, such as conflicting Monthly Recurring Revenue (MRR) figures or runway estimates.
  
• A sophisticated natural language interface, the Nexus AI Chat, allows founders to instantly query the unified knowledge base, summarise metrics for pitch decks, and identify compliance gaps.
  

How I Built It

• Frontend: Built as a Single Page Application using React 18, TypeScript, and Vite. Implemented a custom dark-mode, premium aesthetic using Tailwind CSS 3 and the Nord colour palette. For visual data topology, utilised react-force-graph-2d.

• Backend: Orchestrated via Python 3.12 and FastAPI, managing a local SQLite database using SQLAlchemy.

• AI Integration: All structuring, contradiction detection, and chat features are routed through a custom Ollama service wrapper, interacting directly with local LLMs via REST API on port 11434.

Challenges I Ran Into

• Processing highly unstructured and messy data from multiple simulated sources (Slack messages vs. structured CSVs) required rigorous prompt engineering to force the local LLM to consistently output valid JSON for the structuring engine.

• Managing the physics simulation of the Knowledge Graph to ensure categories and documents separated neatly without visual clutter required fine-tuning D3 force gravity and collision parameters.

Accomplishments I'm Proud Of

• Successfully building a fully functional, zero-exfiltration AI pipeline that guarantees absolute data privacy for sensitive startup financials.

• Developing the “Alignment Checks” contradiction engine, which successfully detects and flags subtle discrepancies between internal operational documents and external investor pitch decks.

What I Learned

• Local LLMs are highly capable of handling complex enterprise data orchestration tasks, demonstrating that startups do not need to compromise their data privacy by relying on external APIs for coherence and structuring.
  

What’s Next for Nexus

• Expanding the suite of autonomous AI agents to not just flag contradictions, but actively draft updated, reconciled documents.

• Integrating real-time webhooks for live data ingestion from production platforms rather than static simulated documents.

Built With

• brain
• lotsofluck
• my
• python