What Phase 1 delivers
Secure Local Node
Oction installs and configures an on-premise inference and retrieval node at your office. Your project files, reports, and calculations never leave Cyntech hardware.
Document Ingestion Engine
Load PDF reports, geotechnical standards, project specs, and historical calculation packages into a searchable project knowledge base.
Engineering Copilot
Ask natural-language questions and receive cited, context-grounded answers with traceability back to the source document and page.
Report Draft Assistant
Generate first-pass sections (site conditions, design rationale, appendices) in Cyntech’s format, ready for engineer review and sign-off.
Hallucination Controls
Built-in source citation, confidence scoring, and a mandatory human-in-the-loop checkpoint before any output is used in client deliverables.
4-Week Validation Plan
Week-by-week milestones, acceptance tests, and a Week 4 benchmark comparing output quality, speed, and cost against current manual workflows.
Commercial model
One-time hardware ~$1,700 (Mac Mini-class local node)
No retainer. No recurring subscription.
Payment terms: 50% on signature, 50% on Week 4 acceptance. If the benchmark does not show measurable time savings and engineer confidence, the second milestone can be withheld until the gap is closed.
Why no cloud-token pricing?
Because the workload runs on the local node, Cyntech pays for compute once (hardware) rather than per-query token costs. Based on the usage profile discussed, this removes an estimated $22,000+ CAD in annual cloud-token spend at scale.
Hours-saved business case
| Use case | Current time | Phase 1 time | Hours saved / report |
|---|---|---|---|
| Report section drafting | 8–12 hrs | 2–4 hrs | 6–8 hrs |
| Standard / spec lookup | 1–2 hrs | 5–15 min | 0.75–1.5 hrs |
| Historical project cross-check | 2–4 hrs | 15–30 min | 1.5–3.5 hrs |
| Client Q&A response prep | 1–3 hrs | 20–40 min | 0.7–2.3 hrs |
Conservative estimate: 10–15 hours saved per major report at full build. At a loaded engineering rate of $120–$160/hr, that is roughly $1,200–$2,400 in direct cost avoidance per report, with faster client turnaround as a secondary benefit.
4-week timeline
Week 1
Deploy & ingest. Install the local node, connect file stores, and ingest a representative set of reports and standards.
Week 2
Retrieval tuning. Build the project knowledge base, test queries, and tune citation formatting to match Cyntech report standards.
Week 3
Copilot & draft workflows. Enable report-section generation, engineer review checkpoints, and confidence scoring.
Week 4
Benchmark & handoff. Run side-by-side comparison against current workflow, document results, and define Phase 2 scope.
Technical guardrails
- Local-only inference: models and data stay inside the Cyntech network on the deployed node.
- Source citation: every answer references the originating document and section.
- Confidence scoring: low-confidence answers are flagged and routed to a human engineer.
- Human-in-the-loop: no AI-generated content goes to clients without engineer review and sign-off.
- Audit trail: all queries and outputs are logged for quality review and liability traceability.
Next step
Sign the Phase 1 proposal and send the deposit. We start Week 1 immediately. If helpful, we will walk Luis and Calvin through the deliverables and hallucination controls on a 30-minute call before signature.