Relay
Internal agent fleet for a logistics SaaS — ticket triage, ERP lookups, exception routing, full audit trails, and human approval on high-risk actions.
- ·tool protocols
- ·HITL
- ·queues
- ·audit log
−62% mean handle time
● orbitone · systems
Agents, retrieval stacks, and AI product surfaces with a model path, a UI, and a number. Some details anonymized for clients.
Internal agent fleet for a logistics SaaS — ticket triage, ERP lookups, exception routing, full audit trails, and human approval on high-risk actions.
−62% mean handle time
Contract Q&A over 40k legal documents. Citation-first answers, continuous eval against partner rubrics, hybrid search + re-rank.
91% citation accuracy
Generation studio for a design tool — multi-model routing, streaming previews, versioning, and rate-aware fallbacks when providers throttle.
3.4× session depth
Tier-1 support copilot with policy-grounded replies, escalation graphs, and red-team tests for prompt injection on customer data.
48% auto-resolve rate
Custom eval harness + CI gates for an AI search product. Offline golden sets, online shadow traffic scoring, cost-per-query dashboards.
evals in every PR
Ingestion and chunking stack for a knowledge product — OCR, structure-aware splits, embedding refresh jobs, and stale-doc detection.
in progress
// your system next
Most builds start as a narrow vertical. Tell us the surface, the data, and the failure modes you care about.