(examples revamp): add example walkthroughs

examples/drive-thru/README.md

@@ -3,3 +3,49 @@
 A complete drive-thru ordering system demonstrating interactive voice agents for food ordering with database integration and order management.
 
 For setup instructions and more details, see the [main examples README](../README.md).
+
+## Overview
+
+This example simulates a fast food drive-thru. It is split across three files: `database.py` contains the menu and formats it as system prompt text, `order.py` holds Pydantic models for the three order types, and `agent.py` defines `DriveThruAgent` with dynamically built ordering tools.
+
+The full menu is loaded once per session and injected directly into the agent's instructions, so the LLM has menu context without needing to call a tool.
+
+### Menu Loading
+
+At the start of each session, `new_userdata()` queries `FakeDB` for all item categories (drinks, combos, Happy Meals, regulars, sauces) and stores them in the `Userdata` dataclass alongside a fresh `OrderState`.
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/agent.py#L382-L399
+`DriveThruAgent.__init__` then formats each category using `menu_instructions()` and concatenates the results with `COMMON_INSTRUCTIONS` to build the full system prompt. This means the LLM sees the entire menu from the first turn and can answer questions or suggest items without any tool calls.
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/agent.py#L55-L83
+### Dynamic Tool Building
+
+The three ordering tools are constructed by `build_combo_order_tool`, `build_happy_order_tool`, and `build_regular_order_tool`. Each method closes over the relevant item lists and injects their IDs as the `enum` constraint in the tool's JSON schema.
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/agent.py#L85-L119
+
+This restricts the LLM to known IDs at the schema layer before any runtime logic runs. `ToolError` handles the cases that can't be caught statically — for example, when a drink has multiple available sizes and the customer hasn't specified one yet, the tool raises a `ToolError` prompting the agent to ask for clarification before retrying.
+
+### Order Types
+
+`order.py` defines three Pydantic models: `OrderedCombo`, `OrderedHappy`, and `OrderedRegular` . A discriminated union `OrderedItem` is also defined. Each ordered item receives a random short `order_id` on creation via `order_uid()`. 
+
+`OrderState` stores the current cart as a `dict[str, OrderedItem]` keyed by `order_id`, which the `remove_order_item` and `list_order_items` tools use to look up or modify existing items.
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/order.py#L45-L56
+
+### Managing the Order
+
+Two tools handle cart management:
+
+- `list_order_items` returns all current cart items with their `order_id`s. The agent is instructed to call this first when modifying or removing an item whose `order_id` is unknown.
+- `remove_order_item` removes one or more items by `order_id`. Modifications (e.g., upsizing fries) are done by removing the old item and re-adding it with the new parameters.
+
+`max_tool_steps=10` is set on the session to give the agent enough budget to call `list_order_items` followed by `remove_order_item` in a single turn when needed.
+
+### Background Audio
+
+`BackgroundAudioPlayer` plays an ambient drive-thru noise track (`bg_noise.mp3`) throughout the session to set the scene. 
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/agent.py#L438-L443
+
+### STT Tuning
+
+The STT model is also initialized with `keyterm` hints for McDonald's brand names (e.g., `"Big Mac"`, `"McFlurry"`, `"McCrispy"`) to improve transcription accuracy.
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/drive-thru/agent.py#L415-L430

examples/frontdesk/README.md

@@ -3,3 +3,40 @@
 A front desk agent demonstrating customer service with calendar integration and appointment management.
 
 For setup instructions and more details, see the [main examples README](../README.md).
+
+## Overview
+
+In this example, you will be able to schedule appointments (optionally with cal.com's API if `CAL_API_KEY` is set) and evaluate the agent's performance using `JudgeGroup`. The session will always begin with the agent saying "Hello, I can help you schedule an appointment."
+
+### Scheduling appointments 
+
+The LLM will call list_available_slots before `schedule_appointment`, since `slot_id` is a required argument. 
+
+`list_available_slots` will return slots like:
+
+```bash
+ST_abc123 - Saturday, January 1, 2000 at 14:00 PDT (in 5 days)
+```
+
+The slots are also cached as a lookup table for `schedule_appointment`. 
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/frontdesk/frontdesk_agent.py#L184
+
+
+This prevents the LLM from passing a hallucinated answer.
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/frontdesk/frontdesk_agent.py#L94-L95
+
+
+The user's email is then collected via `GetEmailTask()`. If the agent is interrupted after the task completes, `schedule_appointment` is aborted before an API call is made to book the slot. After the task, the function is uninterruptible. 
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/frontdesk/frontdesk_agent.py#L97-L119
+
+
+### Evaluations
+
+After the session ends, we use a `JudgeGroup` with pre-built judges to score the conversation. 
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/frontdesk/frontdesk_agent.py#L200-L214
+
+When the success criteria for an agent is clear, using judges can complete the evaluation by measuring the performance quality. 

examples/frontdesk/frontdesk_agent.py

@@ -77,7 +77,7 @@ def __init__(self, *, timezone: str) -> None:
         self._slots_map: dict[str, AvailableSlot] = {}
 
     async def on_enter(self) -> None:
-        await self.session.say("hello, I can help you to schedule an appointment")
+        await self.session.say("Hello, I can help you schedule an appointment!")
 
     @function_tool
     async def schedule_appointment(

examples/healthcare/README.md

@@ -0,0 +1,46 @@
+# Healthcare Example
+
+A full healhcare assistant providing secure appointment management, billing handling, and lab result retrieval.
+
+For setup instructions and more details, see the [main examples README](https://github.com/livekit/agents/blob/main/examples/README.md).
+
+## Overview
+
+The healthcare agent utilizes a variety of `AgentTasks` to achieve structured workflows to collect information. This example is modality-agnostic, where users can interact via text or voice and switch seamlessly. If the conversation heads out of the scope of the agent, the user will be transfered to a human.
+
+### Profile Authentication
+
+Before any sensitive information is queried, the user must go through an authentication process. This process will only occur once per call. If the user provides a name and birthday existing in the database, the process is fast-forwarded. This is possible via task completion callbacks:
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L574-L588
+Otherwise, the user will also be asked for their phone number and insurance provider
+to create a profile.
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L590-L641
+
+After the profile is created, the agent will be given a tool to update the patient's record as needed.
+
+### Appointment Management
+
+Function tools are added dynamically, so the LLM cannot hallucinate parameters or call tools prematurely. The user is given options for compatible doctors based on their insurance:
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L306-L333
+
+After the doctor is chosen, the appointment scheduling tool is built dynamically with the availabilities. 
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L734-L780
+
+Finally, the visit reason is collected, and once confirmed the database will be updated accordingly (the doctor's availability will be removed).
+
+Users are also able to modify existing appointments. If the user wishes to reschedule an appointment, the appointment is canceled and `ScheduleAppointmentTask` is reused. 
+
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L506-L543
+
+### Billing Handling
+
+`GetCreditCardTask()` is showcased here. The user's details are verified, and a balance is generated and connected to their profile.
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L734-L748
+
+### Lab Result Retrieval
+
+We use OpenAI's provider tool, `FileSearch`, to read through an uploaded lab report (mock_checkup_report.pdf). 
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/healthcare/healthcare_agent.py#L734-L780
+

examples/survey/README.md

@@ -1,5 +1,58 @@
-# Survey Example
+# Survey Agent
 
-Example showing how to build an automated voice survey agent.
+Screen a candidate for a software engineer role to see if they meet the prerequisites and are an overall good fit. The responses, summary, and evaluation will be written to a CSV file.
 
 For setup instructions and more details, see the [main examples README](../README.md).
+
+## Overview
+
+The flow of this agent is flexibly structured, where the specified sequence is maintained but the user is able to regress to a previously visited task if needed. This is possible via `TaskGroup`, which is set up here: https://github.com/livekit/agents/blob/f8efe436afe2470104ce7587f1d89ae383ed619e/examples/survey/survey_agent.py#L285-L315
+
+### IntroTask
+This stage facilitates introductions and collects the candidate’s name.
+
+### GetEmailTask
+This task is built in to our framework. By default, it can collect and update emails and mark when a user doesn’t want to give their email. If the input modality is audio, emails are confirmed before the task is marked as complete. See the [docs for GetEmailTask](https://docs.livekit.io/agents/prebuilt/tasks/get-email/).
+
+### CommuteTask
+This stage collects whether or not the candidate can commute to the office and their method of transportation. We define the possible commute methods here:
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/survey/survey_agent.py#L32
+
+And we pass this to a function tool like so:
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/survey/survey_agent.py#L231-L237
+
+### ExperienceTask
+This stage collects the candidate’s years of experience and a short description of their professional career. It follows a structure similar to `IntroTask` and `CommuteTask`.
+
+### BehavioralTask
+For some tasks, you might not want a structured flow of questions. In this stage, we are collecting the candidate’s strengths, weaknesses, and work style. This task incrementally collects answers in no particular order. This allows for a more natural conversation.
+
+After the candidate answers one of the questions, `self._check_completion()` is called to check if all 3 fields (`”strengths”`, `“weaknesses”`, `“work_style”`) have been collected. If so, then `BehavioralTask` is marked as complete. If not, then the agent will continue prompting for the rest of the answers.
+
+In practice, this would ensure variability among candidates’ experiences.
+
+### Closing out
+Once the interview is concluded and TaskGroup is completed, we extract the summary message (the last inserted message):
+
+```python
+summary = self.chat_ctx.items[-1]
+```
+
+And we generate a candidate evaluation based off of the summary:
+
+```python
+evaluation = await evaluate_candidate(llm_model=self.session.llm, summary=summary)
+```
+
+The session LLM evaluates the candidate from the given summary:
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/survey/survey_agent.py#L76-L98
+
+
+Finally, the agent hangs up and you can find the results, summary, and evaluation in `results.csv`!
+
+### Disqualification
+In each stage after the first, the candidate may be disqualified for unsatisfactory answers or for refusing to answer. We create a function tool that will be passed to the tasks:
+https://github.com/livekit/agents/blob/8283a5a5c9863a07bcf030ee90e8ab780e1e569b/examples/survey/survey_agent.py#L101-L118
+
+The candidate will be informed of the interview ending, and then the session will shut down.
+