Intermediate · Agents
Workflows and agents Have the help desk do its own work as well as answer: the desk's tools over a caller's tickets and its advisers' callbacks, a workflow that sorts each message and follows a fixed path, an agent that calls the tools itself, a review of an assistant's agent, and a router that sends the workflow what it can follow and the agent the rest.
You build desk.py, then workflow.py, then agent.py, then desk_agent.py, then router.py
Stages 3 Build, 1 Catch, 1 Ship
Time About 3 hours
Runs on Your machine Loading the workspace…
Loading the workspace…
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . The dataset's 488 help pages in passages: each page's text taken from its HTML, with its headings marked by #s and its list items by -, and its blocks joined into passages of up to 200 words. Each passage has its page, the page's domain, and the page's title without the part number the dataset adds.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . Each passage's vector from nomic-embed-text-v1.5 at 8 bits, embedded after `search_document: `, one text a request, by llama.cpp's server, release b11734, each number times 32,767, rounded.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . The help desk's 120 tasks, written for the lab: 40 of the dataset's opening questions from its test dialogues, of seven words or more, 10 for each service, drawn at random, with the passages that hold each one's answer; and 80 conversations about the caller's tickets and callbacks, each with the caller's messages, tickets and the advisers' free times.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . Each task done by the lab's workflow, agent and router, with Qwen3.5-4B at 4 bits and nomic-embed-text-v1.5 at 8 bits, through llama.cpp's server, release b11734, on machines with 4 CPU cores and no GPU, on 21 October 2026: the agent's five runs of each task and the workflow's and the router's three, each with a seed of its own, at the sampling settings Qwen3.5-4B's model card gives for its non-thinking mode, with whether the task was done, the seconds the run took, its requests, their tokens of prompt and of reply, and the tools the model called in each request.
Model: Qwen3.5-4B , by Qwen, Apache License 2.0 .
Model: nomic-embed-text-v1.5 , by Nomic AI, Apache License 2.0 .