Do I need a fine-tuned sentiment model?

No. A general-purpose LLM with a focused prompt produces strong sentiment, emotion, and entity results across many domains. You trade a little latency for adaptability and a much smaller codebase.

Which LLM provider is used?

The workshop repo uses a provider abstraction with OpenRouter as the default (free tier available), plus Fireworks, Gemini, and OpenAI. You only need one API key to follow along.

Is LangGraph required?

No. The agent graph module detects whether LangGraph is installed. If it is, you get a StateGraph. If not, the plain-async fallback runs the same tasks in the same order.

Premium course

Build a sentiment service that adapts the moment product asks for a new label

Name: Sentiment classification with LLMs and few-shot prompting
Price: 49 USD
Availability: InStock

Classic NLP classifiers lock you into the labels they were trained on. LLM-powered tasks let you change labels, add languages, and introduce new emotions in a single prompt edit. You will build the full service: sentiment, entities, keywords, summary, emotion, streamed per task over SSE.

Enroll Preview curriculum

Still deciding? Ask first.

Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.

Curriculum fit, prerequisites, or where to start
Honest answer, no pressure to enroll

Engineers are learning here from

NVIDIAMICROSOFTGRABWISEPIPEDRIVEBOLTGLIA

Build a multi-task NLP service that classifies sentiment, extracts entities and keywords, summarizes content, and detects emotion. Each task is a focused LLM prompt, orchestrated through a registry, optionally wired into a LangGraph StateGraph, and streamed over Server-Sent Events.

Build a multi-task NLP service powered by focused LLM prompts and per-task streaming.

What you'll ship

Real projects, not toy demos.

A FastAPI service that classifies sentiment as positive, negative, or neutral with a polarity score
A task registry that runs entities, keywords, summary, and emotion tasks behind one interface
A LangGraph StateGraph that sequences the tasks, with a plain-async fallback when LangGraph is unavailable
A Server-Sent Events stream that pushes per-task progress to the client as each task finishes
A resilient pipeline that isolates failures so one broken task does not take down the rest
An LLM router that picks the right subset of tasks based on the input text

What you'll learn

You finish able to:

Write focused prompts that return strict JSON for sentiment, entities, keywords, summary, and emotion
Parse flexible LLM output safely so a stray markdown fence never breaks your service
Shape task results into Pydantic response models that callers can depend on
Register tasks behind a single interface so adding a new task is a one-line change
Orchestrate tasks with a LangGraph StateGraph and fall back to plain async when LangGraph is not installed
Stream per-task progress over Server-Sent Events and consume it from a simple client
Let the caller pick a subset of tasks, and let an LLM router pick tasks automatically
Isolate failures per task so a broken emotion classifier never breaks sentiment

Curriculum

From a single sentiment prompt to a streaming, routed NLP agent.

01
Single sentiment task
Wire one classify_sentiment prompt end to end, then tighten the output with few-shot examples
2 lessons
02
Structured JSON output
Lock responses into a Pydantic schema and add a repair pattern that survives fenced, truncated, and prose-wrapped LLM output
2 lessons
03
Task registry
Add entities, keywords, summary, and emotion behind a single registry so new tasks become one-line additions
2 lessons
04
Agent graph with LangGraph
Upgrade the dispatcher into a LangGraph StateGraph, with a plain-async fallback when LangGraph is not installed
2 lessons
05
Server-Sent Events streaming
Push per-task progress events to the client as each task finishes, so long requests feel interactive
2 lessons
06
Selective task picker
Let the caller specify a subset of tasks per request, with safe default fallbacks when the input is missing or invalid
2 lessons
07
Resilient errors
Isolate per-task failures so a broken emotion classifier never takes down sentiment, entities, or summary
2 lessons
08
Router upgrade and recap
Let an LLM pick the applicable tasks for each input, then ship the service and recap the whole system
2 lessons

Who it's for

Is this for you?

Python developers

who are stuck with scikit-learn or spaCy classifiers that cannot adapt to new labels without retraining

Backend engineers

who need to add sentiment, summarization, and entity extraction to an existing product without hiring an ML team

AI engineers

who want a clean pattern for orchestrating many small LLM calls and streaming progress to the UI

FAQ

Common questions.

Do I need a fine-tuned sentiment model?
No. A general-purpose LLM with a focused prompt produces strong sentiment, emotion, and entity results across many domains. You trade a little latency for adaptability and a much smaller codebase.
Which LLM provider is used?
The workshop repo uses a provider abstraction with OpenRouter as the default (free tier available), plus Fireworks, Gemini, and OpenAI. You only need one API key to follow along.
Is LangGraph required?
No. The agent graph module detects whether LangGraph is installed. If it is, you get a StateGraph. If not, the plain-async fallback runs the same tasks in the same order.

Pricing

Pick the path that fits.

Subscribe to Pro for every paid course, or buy just this one.

Subscribe to Pro

Recommended

Unlock this course and every paid course plus workshop replays. One subscription.

$35$16/mo · Billed annually

You save 54% with regional pricing

Every paid course unlocked
All workshop replays included
New releases as they ship
Cancel anytime

Buy this course

One-time purchase. Lifetime access to every lesson, exercise, and update.

$149$79one-time

You save 47% with regional pricing

Lifetime access to every lesson and future update
Hands-on projects you can show in interviews
Certificate when you finish

Still deciding? Ask Param a question

After this course:

NLP tasks are prompts now. Learn the shape once.

Enroll

Sentiment classification with LLMs and few-shot prompting

$79 one-time

Enroll

Premium course

Build a sentiment service that adapts the moment product asks for a new label

Enroll Preview curriculum

Still deciding? Ask first.

Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.

Curriculum fit, prerequisites, or where to start
Honest answer, no pressure to enroll

Engineers are learning here from

NVIDIAMICROSOFTGRABWISEPIPEDRIVEBOLTGLIA

Build a multi-task NLP service powered by focused LLM prompts and per-task streaming.

What you'll ship

Real projects, not toy demos.

A FastAPI service that classifies sentiment as positive, negative, or neutral with a polarity score
A task registry that runs entities, keywords, summary, and emotion tasks behind one interface
A LangGraph StateGraph that sequences the tasks, with a plain-async fallback when LangGraph is unavailable
A Server-Sent Events stream that pushes per-task progress to the client as each task finishes
A resilient pipeline that isolates failures so one broken task does not take down the rest
An LLM router that picks the right subset of tasks based on the input text

What you'll learn

You finish able to:

Write focused prompts that return strict JSON for sentiment, entities, keywords, summary, and emotion
Parse flexible LLM output safely so a stray markdown fence never breaks your service
Shape task results into Pydantic response models that callers can depend on
Register tasks behind a single interface so adding a new task is a one-line change
Orchestrate tasks with a LangGraph StateGraph and fall back to plain async when LangGraph is not installed
Stream per-task progress over Server-Sent Events and consume it from a simple client
Let the caller pick a subset of tasks, and let an LLM router pick tasks automatically
Isolate failures per task so a broken emotion classifier never breaks sentiment

Curriculum

From a single sentiment prompt to a streaming, routed NLP agent.

01
Single sentiment task
Wire one classify_sentiment prompt end to end, then tighten the output with few-shot examples
2 lessons
02
Structured JSON output
Lock responses into a Pydantic schema and add a repair pattern that survives fenced, truncated, and prose-wrapped LLM output
2 lessons
03
Task registry
Add entities, keywords, summary, and emotion behind a single registry so new tasks become one-line additions
2 lessons
04
Agent graph with LangGraph
Upgrade the dispatcher into a LangGraph StateGraph, with a plain-async fallback when LangGraph is not installed
2 lessons
05
Server-Sent Events streaming
Push per-task progress events to the client as each task finishes, so long requests feel interactive
2 lessons
06
Selective task picker
Let the caller specify a subset of tasks per request, with safe default fallbacks when the input is missing or invalid
2 lessons
07
Resilient errors
Isolate per-task failures so a broken emotion classifier never takes down sentiment, entities, or summary
2 lessons
08
Router upgrade and recap
Let an LLM pick the applicable tasks for each input, then ship the service and recap the whole system
2 lessons

Who it's for

Is this for you?

Python developers

who are stuck with scikit-learn or spaCy classifiers that cannot adapt to new labels without retraining

Backend engineers

who need to add sentiment, summarization, and entity extraction to an existing product without hiring an ML team

AI engineers

who want a clean pattern for orchestrating many small LLM calls and streaming progress to the UI

FAQ

Common questions.

Do I need a fine-tuned sentiment model?
No. A general-purpose LLM with a focused prompt produces strong sentiment, emotion, and entity results across many domains. You trade a little latency for adaptability and a much smaller codebase.
Which LLM provider is used?
The workshop repo uses a provider abstraction with OpenRouter as the default (free tier available), plus Fireworks, Gemini, and OpenAI. You only need one API key to follow along.
Is LangGraph required?
No. The agent graph module detects whether LangGraph is installed. If it is, you get a StateGraph. If not, the plain-async fallback runs the same tasks in the same order.

Pricing

Pick the path that fits.

Subscribe to Pro for every paid course, or buy just this one.

Subscribe to Pro

Recommended

Unlock this course and every paid course plus workshop replays. One subscription.

$35$16/mo · Billed annually

You save 54% with regional pricing

Every paid course unlocked
All workshop replays included
New releases as they ship
Cancel anytime

Buy this course

One-time purchase. Lifetime access to every lesson, exercise, and update.

$149$79one-time

You save 47% with regional pricing

Lifetime access to every lesson and future update
Hands-on projects you can show in interviews
Certificate when you finish

Still deciding? Ask Param a question

After this course:

NLP tasks are prompts now. Learn the shape once.

Enroll

Sentiment classification with LLMs and few-shot prompting

$79 one-time

Enroll