Implementation:Truera Trulens TruBot App

Knowledge Sources	Truera_Trulens
Domains	LLM Applications
Last Updated	2026-02-14 08:00 GMT

Overview

A LangChain-based conversational chatbot (TruBot) that uses Pinecone for vector retrieval and OpenAI for question answering, integrated with Slack and instrumented with TruLens feedback functions.

Description

The TruBot application, defined in examples/experimental/end2end_apps/trubot/trubot.py, is a complete end-to-end conversational bot that combines several technologies:

• Pinecone -- Used as a vector database for document retrieval. The bot connects to an existing Pinecone index named "llmdemo" using OpenAI embeddings (text-embedding-ada-002).
• LangChain -- Provides the ConversationalRetrievalChain framework that ties together the retriever, LLM, and conversation memory into a unified chain.
• OpenAI -- Powers the language model (OpenAI with temperature 0 and 128 max tokens) for generating answers.
• Slack -- The bot operates as a Slack application using slack_bolt, handling direct messages and mentions.
• TruLens -- Instruments the chain via TruChain and evaluates quality using three feedback functions.

The bot supports five app variants (selectable per conversation) that experiment with different prompt strategies:

• 0/default -- The standard LangChain prompt.
• 1/lang_prompt -- Modified prompt that instructs the LLM to respond in the same language as the question.
• 2/relevance_prompt -- Modified prompt that instructs the LLM to use only relevant contexts.
• 3/filtered_context -- Uses TruLens WithFeedbackFilterDocuments to filter retrieved documents below a relevance threshold of 0.5.
• 4/filtered_context_and_lang_prompt -- Combines filtered context with the language-matching prompt.

Three TruLens feedback functions are defined:

• f_lang_match -- Evaluates language match between question and answer using HuggingFace.
• f_qa_relevance -- Evaluates question/answer relevance using OpenAI.
• f_context_relevance -- Evaluates question/context chunk relevance using OpenAI, aggregated with np.min.

Usage

This application is used as an experimental end-to-end demo of TruLens instrumentation on a production-like chatbot. It demonstrates how to set up feedback functions, instrument LangChain chains with TruChain, and compare multiple app variants for conversational quality evaluation. It is also a practical example of integrating TruLens with Slack bots.

Code Reference

Source Location

• Repository: Truera_Trulens
• File: examples/experimental/end2end_apps/trubot/trubot.py
• Lines: 1-378

Signature

def get_or_make_app(
    cid: str,
    selector: int = 0,
    feedback_mode=feedback_schema.FeedbackMode.DEFERRED,
) -> mod_tru_chain.TruChain: ...

def get_answer(
    app: mod_tru_chain.TruChain,
    question: str,
) -> Tuple[str, str]: ...

def answer_message(client, body: dict, logger): ...

def start_bot(): ...

Import

from examples.experimental.end2end_apps.trubot.trubot import get_or_make_app, get_answer

I/O Contract

Function: get_or_make_app

Direction	Name	Type	Description
Input	cid	str	Slack conversation ID (channel ID or thread timestamp).
Input	selector	int	App variant selector (0-4); determines prompt strategy and retrieval filtering.
Input	feedback_mode	FeedbackMode	Feedback evaluation mode; defaults to DEFERRED.
Output	return	TruChain	A TruLens-instrumented LangChain conversational retrieval chain.

Function: get_answer

Direction	Name	Type	Description
Input	app	TruChain	The TruLens-instrumented chain to use for answering.
Input	question	str	The user question to answer.
Output	return	Tuple[str, str]	A tuple of (answer text, sources elaboration text).

Function: answer_message

Direction	Name	Type	Description
Input	client	WebClient	Slack API WebClient instance for posting messages.
Input	body	dict	Slack event body containing message details (text, channel, user, timestamp).
Input	logger	Logger	Logger instance for recording event activity.
Output	(side effect)	--	Posts answer and sources back to the Slack channel/thread.

Required Environment Variables

Variable	Description
OPENAI_API_KEY	API key for OpenAI services.
HUGGINGFACE_API_KEY	API key for HuggingFace inference.
PINECONE_API_KEY	API key for Pinecone vector database.
PINECONE_ENV	Pinecone environment identifier.
SLACK_TOKEN	Slack bot OAuth token.
SLACK_SIGNING_SECRET	Slack app signing secret for request verification.

App Variants

Selector	Version	Description
0	0/default	Standard LangChain default prompt.
1	1/lang_prompt	Prompt instructs LLM to respond in same language as the question.
2	2/relevance_prompt	Prompt instructs LLM to use only relevant contexts.
3	3/filtered_context	Retriever filters documents below relevance threshold 0.5.
4	4/filtered_context_and_lang_prompt	Combines filtered context with language-matching prompt.

Usage Examples

from examples.experimental.end2end_apps.trubot.trubot import get_or_make_app, get_answer

# Create or retrieve a TruChain app for a given conversation
tc = get_or_make_app(cid="C0123456789", selector=1)

# Get an answer from the instrumented chain
answer_text, sources_text = get_answer(tc, "What is TruLens?")
print(answer_text)
print(sources_text)

To run the full Slack bot:

# Set required environment variables, then:
python examples/experimental/end2end_apps/trubot/trubot.py
# Bot starts on port 3000 and listens for Slack events

Related Pages

• Environment:Truera_Trulens_Python_Core_Environment