Kalsnet (KNet) Consulting Group | AI • Cyber • Agentic Systems/GraphTheory
GraphTheory is a single-file Streamlit application that turns thirteen defense and cybersecurity analysis scenarios — attack path mapping, MITRE ATT&CK correlation, identity and access risk, threat intel fusion, insider threat detection, software supply chain risk, intelligence investigation, mission readiness, logistics resilience, counter-terrorism financing, personnel security, infrastructure digital twins, and data sovereignty — into interactive knowledge graphs. For each use case it either generates realistic synthetic node and edge data or accepts uploaded CSVs, builds a directed graph, and computes structural metrics like degree centrality, betweenness, and PageRank to surface the most important nodes and shortest paths (e.g., an attacker's route from a compromised endpoint to a critical asset). The underlying tools are Streamlit for the UI, NetworkX for graph construction and analytics, Plotly for interactive network and bar-chart visualizations, Graphviz for schema flow diagrams, and python-docx/ReportLab for exporting findings to Word and PDF. AI's role is interpretive rather than computational: once the graph metrics are computed, a summarized structural snapshot (top-ranked nodes, relationships, totals) is sent to an LLM — Groq or Gemini, selectable in the sidebar — which translates that raw graph data into a plain-language narrative: explaining attacker reachability, attributing activity to a threat group, flagging escalation paths, or recommending remediation priority, always grounded in the supplied metrics rather than invented facts. Real-world benefit is turning fragmented, hard-to-read connected data into a defensible, explainable analyst workflow: instead of scanning isolated alerts or spreadsheets, a security or intelligence team gets a visual attack/impact path, a ranked list of what matters, and an AI-written explanation they can act on or hand to leadership. The export tab then packages the graph, metrics, and AI narrative into a shareable PDF, Word, text, or CSV report for briefings, audits, or case files — collapsing what would normally be a multi-tool investigation (SIEM, spreadsheet, graph database, report writer) into one guided .
Launch AppHow GraphDB Studio works: Your data (a CSV of relationship records) gets loaded into an in-memory graph using NetworkX (a Python graph library) — turning rows into nodes and edges. Four tabs run different graph operations on that structure: rendering the full network, finding the shortest path between two nodes, computing centrality (who’s most connected/influential), and aggregating relationship stats. Each result renders as an interactive Plotly chart/graph, plus a table you can export. Role of AI None, really — no AI or LLM is involved in the app’s logic. It’s classical graph algorithms (traversal, centrality math) via NetworkX. I (Claude) was used to write the code, not to run inside it. Tools/libraries used: Streamlit (UI/GUI) NetworkX (graph modeling & algorithms) Plotly (interactive charts/network viz) pandas (data handling) python-docx / reportlab (Word/PDF export) Real-world benefits: Graph databases (Neo4j, Neptune, etc.) shine where relationships matter more than tables: fraud detection, recommendation engines, social network analysis, org/knowledge graphs, supply-chain mapping. This app is a sandbox to prototype that thinking with your own data before committing to a real graph DB — and the same node/edge model transfers directly if you later move to one. .
Launch App