What DaVinci has delivered as a firm.

Business problem

Margins in import distribution are set by two things: how long inventory sits, and how many miles run empty. Wayfong could measure neither in time to act. Buyers ordered from experience rather than forecast, warehouse staff searched for pallets by memory, and dispatch built routes by hand each morning. Growth was making all three worse.

Scope DaVinci owned

End-to-end: discovery, target architecture, the full build, data migration off the legacy system, cutover, training and post-launch stabilization. DaVinci was accountable for delivery of the operating platform. Wayfong retained ownership of commercial policy, supplier relationships and warehouse process design, which we built to rather than replaced.

Starting environment

A legacy on-premise inventory and order system with no API surface, a separate accounting package, and substantial operational knowledge held in spreadsheets and in people’s heads. Cold-chain product with hard shelf-life constraints, multi-temperature warehousing, and customs and food-safety documentation requirements on inbound containers.

Solution

A cloud-native operating platform with a multi-agent AI layer above the operational core. In business terms: the system now tells buyers what to order and when, tells warehouse staff exactly where to go and in what sequence, and builds the next day’s routes before anyone arrives.

• AI purchase officer — procurement recommendations and inventory forecasting against demand, lead time and shelf life
• Logistics planner — automated route construction and dispatch sequencing
• Spare capacity auction agent — dynamic pricing of unused truck space
• Inventory management, warehouse mapping and slotting, order-to-cash workflow, supplier and PO management, and executive reporting

Delivery shape

A principal sponsor accountable for architecture and executive reporting, one technical lead, an engineering pod of six to eight, and shared QA and platform support. Two-week sprints with a demo every sprint. Phased cutover by function — inventory first, then purchasing, then dispatch — so the business never sat on a hard switchover.

Technology

React and TypeScript on the front end, Node.js services, PostgreSQL, Python for forecasting and routing models, deployed on AWS with managed CI/CD. Integrations to the incumbent accounting package and to carrier and customs data sources.

Business problem

KYC refresh and transaction monitoring alerts were queuing. Each case required an analyst to pull documents from four systems, read them, and write a narrative — work that was slow, inconsistent between analysts, and hard to defend when a regulator sampled it months later.

Scope DaVinci owned

Solution architecture, the retrieval and evaluation layer, the analyst workflow application, and the audit and logging model. The bank owned case policy, escalation thresholds and the model risk approval process. We built to their control framework rather than proposing our own.

Starting environment

Case management, core banking, document storage and a sanctions screening tool, none of which shared an identity model. Regulated data that could not leave the bank’s cloud tenancy, an internal model risk management review gate, and an existing security review process every component had to clear.

Solution

A review layer that retrieves the relevant case material, drafts a structured assessment with every assertion linked to the source document it came from, and routes to a human for disposition.

• Retrieval grounded in bank-held documents only, with citation back to source
• Deterministic escalation rules outside the model — thresholds are code, not prompt
• Full audit trail: inputs, retrieved context, model version, output, reviewer, decision
• Human-in-the-loop disposition — the system never closes a case

Delivery shape

Principal sponsor, technical lead, an engineering pod of five, and a dedicated security engineer through the review gates. Delivered inside the bank’s own cloud tenancy and Git organization from sprint one.

Technology

Client-tenanted cloud deployment, vector retrieval over bank-held documents, enterprise model endpoints under the bank’s own provider agreement, Python services and a React analyst interface.

Business problem

Yard allocation, carrier routing and exception handling were decided by experienced planners under time pressure. The decisions were good but unrepeatable, they did not scale with volume, and the institutional knowledge behind them sat with a small number of people.

Scope DaVinci owned

The agent framework, the optimization models, integration into the terminal operating system and customs interfaces, and the planner-facing application. The authority retained operational policy and final override authority at every decision point.

Starting environment

An established terminal operating system with a constrained integration surface, customs and manifest interfaces with fixed data contracts, and a hard requirement that the new system degrade to manual planning without interrupting terminal operations.

Solution

Agents scoped to the decisions planners were already making, with the planner kept in the loop and able to override.

• Yard allocation agent working against dwell time, equipment position and vessel schedule
• Carrier routing and appointment sequencing
• Exception handling with escalation to a human planner on defined conditions
• Planner console showing every recommendation and the reasoning inputs behind it

Delivery shape

Principal sponsor, technical lead, an engineering pod of seven, and an operations research specialist on the optimization models. Deployed in shadow mode for a full quarter before taking any live decision.

Technology

Cloud-native services, Python optimization and agent orchestration, event streaming off the terminal operating system, and a React planner interface.