Talk to your data.
Get answers, not queries.
SVOD turns your relational data into a knowledge graph, then lets your team ask plain-English questions and get executive-ready answers — Cypher, evidence, confidence, and all.
1,000 accounts · 273K relationships · ~4 LLM calls per query
Features
An agent that shows its work.
Schema-locked Cypher
Every generated query is read-only and safety-filtered. Internal Neo4j IDs and unsafe operations are rejected before they reach the database.
Self-correcting agent
Failed queries are repaired automatically — the LLM sees the exact Neo4j error and produces a fixed Cypher in one shot.
Evidence-backed reasoning
For 'why' questions the agent plans 2-3 supporting queries, validates each, and synthesises an answer grounded in real rows.
Deep Think mode
Toggle iterative analysis — up to four cycles of plan → run → reflect. Each cycle adds a new angle until the answer holds.
LLM-as-judge confidence
Every answer is scored across relevance, completeness, grounding, and sufficiency. Low-confidence results can escalate to Deep Think with one click.
Interactive visualizations
Bar, line, area, pie, and scatter charts on every result set. Pick your axes and chart type — built on Recharts.
Live progress streaming
Server-Sent Events push agent state to the UI — see 'Planning…', 'Cycle 2/4 — reflecting…', and per-query progress as it happens.
Multi-turn memory
Follow-up questions resolve referents like 'those users' and 'each plan' using prior turns. Suggested next questions follow every answer.
How it works
Three steps from data to insight.
Bring your data
We ingest any relational source and reshape it into a Neo4j knowledge graph — accounts, services, content, relationships.
Ask anything
Business users type plain English. The agent classifies intent, generates schema-locked Cypher, and self-corrects on failure.
Get answers
Executive summary, key metrics, insights, and confidence — with evidence queries and a Deep Think mode for causal questions.
The demo dataset
SVOD streaming customer data.
A sampled subscription-video-on-demand dataset modelled as a Neo4j knowledge graph. Three CSVs (accounts, services, viewership) become a six-label graph the assistant can traverse in one query.
sampled customer accounts
subscription instances across 4 plans
episodes & features (FEATURE / EPISODE)
Drama, Comedy, Thriller, Horror, etc.
monthly buckets for acquisitions, churn, views
SUBSCRIBED_TO, VIEWED, CHURNED_IN, HAS_GENRE…
What you can ask
Built on
A modern, deployable stack.
See it answer your questions
in real time.
Open the assistant and ask anything about the knowledge graph. Cypher, evidence, and confidence visible on every answer.