say hello

ResolveOps

AI support workflow · Jul 2026 evidence pass · Local project

An operator-facing D2C support dashboard that classifies tickets, retrieves order and policy context, proposes safe resolutions, drafts replies, and waits for human approval.

All work

What it is

ResolveOps helps support agents handle order issues without handing decisions to an autonomous bot.

The product promise is practical: collect order, policy, shipment, product, and prior-ticket context; propose a resolution; let the agent approve, edit, escalate, or reject it.

The loop it has to make believable

  1. 01

    Built an agent dashboard around real support operations rather than a customer-facing chatbot.

  2. 02

    Used Redis jobs so analysis is async and auditable instead of blocking the ticket page.

How it’s built

The backend separates ingestion, async analysis, retrieval, rule execution, proposal storage, human decisions, audit logs, monitoring, and evals.

The key design choice is that deterministic rules sit between the model-like classifier and any business action.

Architecture

Data model
Tenants, users, customers, products, variants, orders, shipments, tickets, policies, audit tables
Workflow
Classify, retrieve context, apply deterministic rules, propose action, draft reply, require approval
Safety
No autonomous refunds; risky delivery/refund/damaged-item cases are blocked or escalated
Monitoring
Latency, cost, fallback rate, confidence, approval/rejection/escalation, intent mix, eval summary

What has to hold up

  • Combined retrieval with deterministic refund/return/exchange/delivery rules before a human approval step.
  • Added audit logs, monitoring payloads, five polished demo tickets, and a 20-case golden eval suite.

How I would present it

I would walk through one ticket, show the retrieved citations and rule output, then approve or edit the reply.

The engineering discussion should focus on safe AI workflows: async jobs, full-text retrieval, policy citations, rule gates, audit trails, and evaluation cases.

Proof

ResolveOps is valuable because it frames AI as workflow infrastructure: retrieval, deterministic guardrails, human approval, auditability, and evals.

VerifyRepository, checks, stack, status
  1. 01

    Start with the project status and available evidence.

  2. 02

    Read the architecture and edge-case sections.

  3. 03

    Follow the documented proof path: README, schema/seed SQL, workflow code, eval endpoint, monitoring payload.

Role
Workflow design, Go API/worker, rule engine, dashboard, evals
Scope
Operator console for AI-assisted order support
Proof
README, schema/seed SQL, workflow code, eval endpoint, monitoring payload
Stack
Go, Next.js, PostgreSQL, Redis, Docker Compose, Full-text search, TypeScript
Golden cases
20
Queue
Redis
Backend
Go
Decisioning
Human gated