v3.4 — live evaluation dashboard just shipped

Sage

Reads your documents, answers with cited sources, and takes the next action — no human needed to close the loop.

Built by EFFOMA. A RAG + multi-agent pipeline reused across 5+ enterprise engagements since 2023.

Try it live See the architecture
81%
recall@5, self-hosted embeddings
BGE-large, Vietnamese legal corpus · 45ms, ~$0 marginal cost
~8 sec
30-page contract analysis
vs ~2hr manual review · multi-agent orchestration
~$350/mo
steady-state cost
10K docs, ~1K queries/day · S3, Lambda, OpenSearch, Bedrock
70%
inference cost cut
ElastiCache semantic caching, repeated queries
Architecture

Every default is a decision, not a shrug

Six components, six explicit trade-offs — the answer we give clients who ask "why this and not that."

Sage architecture diagram: ingestion pipeline, supervisor agent orchestration, and evaluation loop with trade-off annotations
01

Ingestion Pipeline

Trade-off Step Functions over Lambda chains: per-step retry and observability, more infra. Breaks down at Glue-scale volume.
02

Vector Store

Trade-off OpenSearch Serverless has a ~$200/mo OCU floor no matter the volume. Light workloads: use pgvector instead.
03

Agent Runtime

Trade-off Chosen per engagement, up front. Defer it and you re-architect memory + guardrails later — no neutral middle ground.
04

Multi-Agent Topology

Trade-off Supervisor adds +1 hop of latency, stays auditable. Swarm prototypes faster but can't explain its own decisions.
05

Evaluation Pipeline

Trade-off Triage by failure type: retrieval miss → chunking/embedding, wrong answer → prompt/model, stale answer → sync pipeline.
06

Security & Access Control

Trade-off Access filtering happens at retrieval time, never post-generation — the only defensible design in a regulated environment.
Live demo

Try it live

Real deployed endpoints, not a mock. Ask the live RAG pipeline, or run the golden-set evaluation suite.

Ask Sage

POST /ask

Evaluation Suite

GET /evaluate

Golden-set retrieval + faithfulness benchmark, gates every release.

Changelog

Release history

Versioned internal platform, shipping since 2024.

v3.4 Latest July 2026
  • Live evaluation dashboard: recall@5, faithfulness, per-question triage
  • Embedding benchmark: BGE-large vs OpenAI ada-002 vs Cohere, Vietnamese legal corpus
  • Now reused across 5+ client engagements
v3.0 Early 2026

AgentCore migration

  • Migrated agent runtime to Bedrock AgentCore (managed memory + guardrails)
  • Added supervisor and hierarchical topology support
  • Vietnamese conglomerate rollout: 5 subsidiaries, 50K docs, 10K queries/day
v2.5 Mid 2025

Evaluation & caching

  • RAGAS integration for automated faithfulness scoring
  • Golden-dataset evaluation harness (100–200 Q&A pairs)
  • ElastiCache semantic caching: ~70% cut in repeated-query inference cost
v2.0 Early 2025

Agentic layer

  • Bedrock Agents orchestrator + specialists (Extract / Compare / Risk / Summary)
  • 30-page contract analysis: ~8 sec vs ~2hr manual review
  • Action-taking added: ticket creation, system lookups
v1.5 Late 2024

Multi-tenant

  • Cognito-based tenant isolation, per-tenant OpenSearch index routing
  • Australia engagement: 7 source systems, 60% scanned/image PDFs, 6 weeks
v1.0 Mid 2024

Initial release

  • Single-tenant RAG pipeline, OpenSearch-backed retrieval
  • S3 → Textract → chunk → embed ingestion flow
  • Basic Q&A over internal documents