Helio
An enterprise knowledge agent based on a knowledge graph that answers organizational questions about service ownership, dependency impact, and decision provenance by traversing the graph, emphasizing visible and verifiable paths.
Timeline (last 90 days)
Helio is a knowledge-graph-based enterprise knowledge agent designed to answer questions about service dependencies, team ownership and decision provenance.
The Helio team used FalkorDB to build a synthetic fintech knowledge graph with 7 node types and 14 edge types, simulating an organization of 120 people during a payments migration.
Before generating any data, the Helio team hand-wrote 20 facts about the company in GROUND_TRUTH.md as the answer key, of which 5 questions have no answer in the graph.
In Helio's scoring framework, refusing to answer is set as an explicit scoring category with the same weight as a correct answer, which changes the objective of prompt optimization.
Using only ORDER BY count DESC returns Priya Sharma, who has the most commits, which is technically correct in quantity but completely wrong as an answer to 'who knows this service' because it ignores recency.
Helio's solution is to return two independent signals, count and recency, and refuse to blend them, for example noting that the platform team owns the payments service, Ravi Menon was still working on it recently (2026-09-28), and Priya Sharma has the most commits but is out of date.
Helio's validation gate once wrongly judged four correct refusals as failures, because it required eight spellings of 'no' to appear in the same sentence (AND logic).
All Facts (7)
Helio is a knowledge-graph-based enterprise knowledge agent designed to answer questions about service dependencies, team ownership and decision provenance.
50%UnverifiedThe Helio team used FalkorDB to build a synthetic fintech knowledge graph with 7 node types and 14 edge types, simulating an organization of 120 people during a payments migration.
50%UnverifiedBefore generating any data, the Helio team hand-wrote 20 facts about the company in GROUND_TRUTH.md as the answer key, of which 5 questions have no answer in the graph.
50%UnverifiedIn Helio's scoring framework, refusing to answer is set as an explicit scoring category with the same weight as a correct answer, which changes the objective of prompt optimization.
50%UnverifiedUsing only ORDER BY count DESC returns Priya Sharma, who has the most commits, which is technically correct in quantity but completely wrong as an answer to 'who knows this service' because it ignores recency.
50%UnverifiedHelio's solution is to return two independent signals, count and recency, and refuse to blend them, for example noting that the platform team owns the payments service, Ravi Menon was still working on it recently (2026-09-28), and Priya Sharma has the most commits but is out of date.
50%UnverifiedHelio's validation gate once wrongly judged four correct refusals as failures, because it required eight spellings of 'no' to appear in the same sentence (AND logic).
50%