Geonix

Capabilities

Demonstrated production-grade systems. Numbers from real deployments.

TB-scale GPS map-matching on commodity hardware

Geospatial data engineering · Production-ready · Architect & lead engineer

  • Industry-standard probabilistic map-matching with spatial indexing for sub-millisecond candidate search
  • Streaming columnar ingestion with smart pre-filtering that discards the bulk of irrelevant input before matching
  • All-cores parallel processing; memory-bound workloads stream without full materialization
  • Benchmark on a 10M+ record cohort: single-digit minutes — substantially faster than common alternatives on the same hardware

Bilingual natural-language query: verified templates + LLM fallback, 11 product surfaces

AI-powered query interfaces · Production-ready · Architect & lead engineer

  • Tiered routing: instant lookups and pre-tested templates first, with an LLM-generated fallback behind validation guardrails
  • Bilingual EN + JP natural-language input over a shared semantic layer
  • Domain-specific safety guardrails: read-only access, bounded results, whitelisted operations
  • Sub-50ms response for the majority of queries

Domain-specific agent orchestration with per-run cost + latency telemetry

Agentic AI for domain systems · Production-ready · Architect & lead engineer

  • Standard tool-server protocol exposing domain capabilities (queries, scoring, validation) to any agent runtime
  • Structured tool I/O with retries, traces, and circuit-breaker patterns to bound failure modes
  • Golden + adversarial eval sets gate every deploy; behavior regressions caught before rollout
  • Per-run cost and latency telemetry wired through standard observability tooling

City-scale traffic microsimulation with demand-calibration capability at Tokyo + Osaka scale

Traffic simulation & synthetic data · Production-ready · Architect & lead engineer

  • Open-source microsimulator (SUMO, EPL-2.0) integrated into the broader analytics pipeline as a synthetic data source
  • Network builds from OpenStreetMap extracts; tested at Tokyo + Osaka city scale
  • Demand calibration against real probe traces, count data, or OD matrices
  • Synthetic trajectory output feeds directly into map-matching, OD extraction, and downstream analytics

Generative interfaces: one natural-language prompt → a validated dashboard (Curator) or multi-layer map (Cartographer)

Constrained generative UI / server-driven UI · Newest extension · Architect & lead engineer

  • One constrained-generative engine, two surfaces: Curator composes multi-widget dashboards (KPI / chart / table / map); Cartographer composes multi-layer maps (hexagon choropleths, flow arcs, hotspots)
  • A self-checking generate-then-validate loop emits only vetted, safe UI — the model cannot invent unsafe components, SQL, or geometry; anything that can't be made safe degrades to a visible placeholder
  • Renders deterministically client-side from the validated spec; data binds to the same guardrailed natural-language query and spatial endpoints
  • Runs behind the same zero-trust boundary (JWT-gated) as the rest of the platform — no public, ungoverned LLM surface

Productized verticals

Nine representative examples from the platform.

Plus 2 more verticals across infrastructure inspection and freight optimization.