Product analytics + LLM tool use

Tidemark: Customer Success Aggregator

A Next.js operating view over the signals a customer-success team touches every day: warehouse-backed health, CRM ownership and contracts, support tickets, engineering issues, usage trends, and a Claude chat sidebar with tool access to the full account context.

Anonymized professional-system rebuild

What is preserved: the system architecture, account-snapshot model, Claude tool-use loop, adapter boundaries, operating decisions, and production deployment shape.

What is fictionalized: the company identity, customer records, source-system details, and domain-specific business logic. The public demo runs entirely on deterministic fixtures.

Build Shape

Canonical Account Snapshot

Combines health, ownership, contract, ticket, engineering, and usage signals behind one stable account shape used throughout the UI.

Operational Next.js Surface

Provides a filterable account book, detail views, health sub-scores, trends, support context, and renewal signals without hiding provenance.

Bounded Claude Tool Use

Passes the full account snapshot inline, then exposes explicit tools only for ticket details, engineering issues, and time-series queries.

Production-Shaped Reference

Includes a reference AWS CDK stack for Fargate, Tailscale, Secrets Manager, and an OIDC deployment role while keeping the demo local.

Engineering Focus

01

One source-of-truth snapshot. The interface and model operate on the same account object, reducing mismatched context across summaries, tools, and decisions.

02

Synthesis, not arithmetic. “Suggest next step” receives pre-computed signals with thresholds already applied, so the model turns evidence into a recommendation instead of inventing business logic.

03

Replaceable data boundaries. Fixture adapters export the same signatures as the production warehouse path, keeping downstream components identical across demo and deployment.