Google: Graph RAG Cuts API Hallucination Rate from 56.4% to 16.2% in Java-to-Python Migration
_reachsumit · x · 2026-09-14
A Google team paper compares Graph RAG vs vector RAG for enterprise Java-to-Python migration, introducing the HCRG (Hierarchical Context-Resident Graph) method.
- Uses tree-sitter for AST extraction, maps architectural edges into Google Cloud Spanner Property Graph, and serializes into Gemini Context Cache for topology-aware, parent-first translation
- Standard RAG fetches isolated chunks that sever inheritance chains; CodeBLEU scored 91% for both methods, masking Standard RAG's syntactically-plausible-but-broken output
- Key gains with Graph RAG: API hallucination drops from 56.4% to 16.2%, Dependency Resolution Quality rises 34.8% → 65.9%, Parent-Child Consistency 26.7% → 45.5%
The team also proposes a custom 7-metric software engineering evaluation framework to replace naive text-overlap metrics.
More from coding & agent
- User builds 20-bot Grok suite to track wins, prep 1:1s and close promises — NicoVerderosa · 2026-10-30
- LangChain CEO opens Interrupt London with keynote on why orgs should own their intelligence — LangChain · 2026-10-30
- Simlock: open-source control plane gives parallel coding agents their own iOS/Android simulators — Scobleizer · 2026-10-30
- Keyfleet launches open canvas where personal AI agents team up with a shared USDC treasury — seanwbren · 2026-10-30
- Page-based project management for Grok Bots: keep chats short, history on Pages — AiJohnAllen · 2026-10-30
- How Oxide Computer uses LLMs internally — FAQ Friday #66 — Sethwinterroth · 2026-10-30