Add LLMClient tutorial and docs

Author: zh1peng

docs/api/psyflow.rst

@@ -52,6 +52,14 @@ psyflow.TriggerSender module
    :undoc-members:
    :show-inheritance:
 
+psyflow.LLM module
+------------------
+
+.. automodule:: psyflow.LLM
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
 psyflow.utils module
 --------------------
 

docs/index.rst

@@ -27,7 +27,8 @@ Start with one of the tutorials below or explore the API reference.
    tutorials/build_stimulus
    tutorials/send_trigger
    tutorials/cli_usage
-   
+   tutorials/llm_client
+
 
 .. toctree::
    :maxdepth: 2

docs/tutorials/llm_client.md

@@ -0,0 +1,92 @@
+# Using `LLMClient` for AI-Driven Helpers
+
+## Overview
+
+`LLMClient` provides a unified interface for interacting with several large language model (LLM) providers and includes helpers for converting tasks to documentation, reconstructing tasks from documentation, and translating experiment resources.
+
+It supports:
+
+- **Google Gemini** via the GenAI SDK (`provider="gemini"`)
+- **OpenAI** models (`provider="openai"`)
+- **Deepseek** models using the OpenAI API format (`provider="deepseek"`)
+
+Custom providers can also be registered programmatically.
+
+## Initialization
+
+Create an instance by specifying the provider, API key, and model name:
+
+```python
+from psyflow import LLMClient
+
+client = LLMClient(
+    provider="openai",
+    api_key="YOUR_API_KEY",
+    model="gpt-3.5-turbo"
+)
+```
+
+The client wraps the underlying SDK and exposes common methods regardless of provider.
+
+## Generating Text
+
+Use `generate(prompt, **kwargs)` to obtain a completion from the configured model. Optional keyword arguments are passed to the underlying provider. Setting `deterministic=True` disables sampling randomness.
+
+```python
+reply = client.generate(
+    "Summarise the Stroop task in one sentence",
+    deterministic=True,
+    max_tokens=50
+)
+print(reply)
+```
+
+## Converting Tasks to Documentation
+
+`task2doc()` summarises an existing task into a README. The function loads your task logic and configuration, optionally uses few‑shot examples from `add_knowledge()`, and returns the generated Markdown text. If `output_path` is provided the README is also written to disk.
+
+```python
+readme_text = client.task2doc(
+    logic_paths=["./src/run_trial.py", "./main.py"],
+    config_paths=["./config/config.yaml"],
+    deterministic=True,
+    output_path="./README.md"
+)
+```
+
+## Recreating Tasks from Documentation
+
+`doc2task()` performs the reverse operation. Given a README or raw description it regenerates the key source files. Provide a directory for outputs via `taps_root` and optionally customise the list of expected file names.
+
+```python
+files = client.doc2task(
+    doc_text="./README.md",
+    taps_root="./recreated_task",
+    deterministic=True
+)
+# files is a dict mapping each file name to its saved path
+```
+
+## Translation Utilities
+
+Several helper methods assist with localisation. The base `translate(text, target_language)` function translates arbitrary text while keeping formatting intact. `translate_config()` applies translation to relevant fields in a psyflow YAML configuration and can write the translated file to disk.
+
+```python
+translated = client.translate(
+    "Press the space bar when you see the target word",
+    target_language="German"
+)
+
+new_cfg = client.translate_config(
+    target_language="German",
+    config="./config/config.yaml",
+    output_dir="./i18n"
+)
+```
+
+These utilities store the last prompt and response for inspection (`last_prompt`, `last_response`) and automatically count tokens for the active model.
+
+## Further Reading
+
+See the API reference for a full description of all attributes and methods provided by [`psyflow.LLMClient`](../api/psyflow#psyflow.LLMClient).
+