Forward Deployed Engineer, how I deliver it, requirement 5 of 12
Request classification and routing
I build automation that classifies each inbound request, plans bounded tasks, and holds high-risk tasks for approval; the dispatcher is in early access.
The requirementBuild automation that reads inbound work, decides what happens next, and acts under human oversight.
How I deliver it at a customer
- Map the inbound flowI sample real requests with the team and agree on the labels, the required fields, and the route for each label. In Luca, an API caller submits a request through the intake app's authenticated POST /api/requests. That call requires the devops:requests:write capability and returns a request id that later stages use.
- Pin the output schemaI write the JSON contract first: intent, kind, risk, complexity, file and token estimates, suggested role, and approval flag. Regex patterns label the obvious requests, and the model is called only when no pattern matches.
- Validate the outputBefore routing, I check the model's answer: a risk other than LOW, MEDIUM, or HIGH becomes MEDIUM, and estimates are clamped to fixed ranges. If the model call fails or no model is reachable, the request falls back to medium risk with human approval required.
- Plan and routeThe dispatch loop picks up each new request, classifies it, and passes it to a planner that writes bounded task envelopes. Each envelope names the agent role, allowed and denied paths, required checks, and a file and line budget. A request that mentions a schema, a migration, or creating or altering a table splits into a schema task and a code task. The code task waits until the schema task is merged. Read-only questions close without opening any task. The planner uses keyword rules today, not a model.
- Gate risky changesEach classifier pattern sets a risk level, and the schema, auth, and deploy patterns set it to high. The planner flags every high-risk task for approval, and a flagged task waits in a state that workers cannot claim. The approve route, POST /api/tasks/:id/approve, requires the devops:dispatch:approve capability and moves the task into the queue.
- Recover and watchThe dispatcher hands out work only when a worker asks for its next task, and it expects heartbeats while the task runs. By default, a task still marked assigned that gets no heartbeat for five minutes goes back in the queue. Its locks on reserved files are released at the same time. The dispatcher dashboard shows the task queue with each task's risk and the locks currently held.
Tools I use
| Tool | What it does here | Status |
|---|---|---|
| Agent dispatcher and plannerEarly access | Classifies each captured request by intent, kind, risk, and role, writes bounded task envelopes, and holds flagged tasks where workers cannot claim them. | Early access |
| Agent tool bridgeAvailable | Exposes database, event, and config services as agent tools, checks the caller's permissions first, and fails a call that runs past its time limit. | Available |
| CRM sales automation engineAvailable | Runs event-triggered workflows with conditional steps and template variables, and stores each step's result, error, and duration. | Available |
| Policy engine approvalsEarly access | Stores pending agent-action approvals with an expiry time and records who decided and when; the dispatcher uses its own approval route. | Early access |
| Demo classifier on Workers AIDemo only | Fills eight fields from the JSON contract my dispatcher's classifier asks its model for, in a public demo environment that calls no Luca system. | Demo only |
Available means shipped and in use. Early access means built but not generally available. Roadmap means designed, not built. Demo only means it powers the public demo, not customer deployments.
Paste an engineering request. The model classifies it into the dispatcher contract: kind, risk, estimates, the agent role it routes to, and whether it must wait for human approval. HIGH risk always requires approval, and output that breaks the contract is rejected.
Open the live demoThe other answers
LLM applications in productionOn Luca, model calls normally go through a gateway that holds provider keys, records each chat call and sets a model per task.Prompt engineeringI write the output shape first, measure the prompt against each model's context window, and check replies against that shape.Retrieval-augmented generationChat answers over a customer's own documents, returned with the matched document names, after a check that the caller may use that persona.AI agents and orchestrationI build agents that call your systems through registered tools, still in early access, and workflows that start on business events.AI workflow orchestrationI map the steps, API contracts, and data flows between a customer's systems and the models before I build the workflow.API and integration designI publish the API contract first, gate it with keys and scopes, then build integrations against it.Hybrid enterprise integrationI connect AI services to your identity, applications and data, and run the platform on Kubernetes on Azure Local.Prototype to productionI take a prototype through generated tests and a two-approval merge gate to a production deploy that rolls back if the new pod keeps crashing or cannot start.Approvals and exception handlingIn chat, I hold each Gatekeeper tool marked destructive until the signed-in user approves it, with expiry and a record. This path is in early access.Evals and output validationI built a versioned rubric judge for a tutoring app, and I trace failed or slow model calls to their provider.Client-facing deliveryI turn engineering work into updates that client architects, security leads and business owners can each act on.
Built by Gus Assuncao of Gus IT LLC for a Forward Deployed Engineer application. Luca is the AI platform Gus IT builds and operates. Contact gus@gusit.de. Claude is a product of Anthropic PBC. Cloudflare and Workers AI are trademarks of Cloudflare, Inc. Llama is a trademark of Meta Platforms, Inc. Other names are trademarks of their respective owners. Gus IT LLC is an independent engineering firm and is not affiliated with or endorsed by these companies.