---

DSPy: Programming—Not Prompting—Language Models

DSPy is a powerful framework designed to shift the paradigm from prompting language models (LMs) with brittle, hard-to-maintain strings to programming structured and modular AI systems. It enables developers to build reliable and scalable AI applications while optimizing prompts and model weights dynamically.

🚀 Why DSPy?

✅ Modular AI System Development

DSPy replaces traditional string-based prompting with structured modules, allowing you to build compositional and optimizable AI pipelines.

🔄 Fast Iteration & Optimization

With DSPy, you can quickly iterate on AI behavior without manually tweaking prompts. It supports self-improving models through integrated optimizers.

🤖 Supports Multiple LMs

Works seamlessly with OpenAI, Anthropic, Databricks, and local models (running on CPU/GPU servers) with simple API configurations.

🏗 From Simple Scripts to Advanced AI Systems

DSPy scales from small proof-of-concept scripts to full-fledged Retrieval-Augmented Generation (RAG), Classification, and Agent-based applications.

---

🛠 Getting Started

🔹 Install DSPy

pip install -U dspy

🔹 Set Up Your Language Model (LM)

import dspy
lm = dspy.LM('openai/gpt-4o-mini', api_key='YOUR_OPENAI_API_KEY')
dspy.configure(lm=lm)

DSPy supports various LMs, including OpenAI, Anthropic, and local LMs. You can authenticate by setting the OPENAI_API_KEY environment variable or passing the API key explicitly.

---

📖 Building DSPy Modules

DSPy allows you to define AI components as structured modules, rather than dealing with unstructured prompt strings. Below are examples of how to build math solvers, RAG systems, classifiers, and agents using DSPy.

🧮 Math Reasoning with DSPy

import dspy
math = dspy.ChainOfThought("question -> answer: float")

result = math(question="Two dice are tossed. What is the probability that the sum equals two?")
print(result)

✅ Possible Output:

Prediction(
    reasoning='When two dice are tossed, each die has 6 faces, resulting in 6 x 6 = 36 possible outcomes...'
    answer=0.0277776
)

---

🔍 Retrieval-Augmented Generation (RAG)

import dspy
from dspy.datasets import HotPotQA

def search_wikipedia(query: str) -> list[str]:
    results = dspy.ColBERTv2(url='http://20.102.90.50:2017/wiki17_abstracts')(query, k=3)
    return [x['text'] for x in results]

rag = dspy.ReAct("question -> answer", tools=[search_wikipedia])
print(rag("What is DSPy?"))

---

📊 Classification with DSPy

class SentimentClassifier(dspy.Module):
    def forward(self, text):
        return dspy.Predict("text -> sentiment: {'positive', 'neutral', 'negative'}")(text)

classifier = SentimentClassifier()
print(classifier.forward("I love using DSPy!"))

---

🎯 Optimizing DSPy Programs

🔹 Self-Improving AI with DSPy Optimizers

DSPy optimizes AI pipelines with built-in optimizers that refine prompts, improve outputs, and even fine-tune weights.

🔥 Example: Optimizing a ReAct Agent

import dspy
from dspy.datasets import HotPotQA

dspy.configure(lm=dspy.LM('openai/gpt-4o-mini'))
trainset = [x.with_inputs('question') for x in HotPotQA(train_seed=2024, train_size=500).train]

react = dspy.ReAct("question -> answer")
tp = dspy.MIPROv2(metric=dspy.evaluate.answer_exact_match, auto="light", num_threads=24)
optimized_react = tp.compile(react, trainset=trainset)

Result: Optimized ReAct performance increased from 24% to 51% accuracy using DSPy MIPROv2 optimizer.

---

🏆 DSPy in Research & Open-Source AI

DSPy emerged from Stanford NLP in 2022, evolving from research in modular LM architectures, inference strategies, and self-optimizing AI. Today, the DSPy community actively contributes to innovations in:

✅ Optimizers: MIPROv2, BetterTogether, LeReT
✅ Architectures: STORM, IReRa, DSPy Assertions
✅ Applications: PAPILLON, PATH, WangLab@MEDIQA, UMD's Prompting Study

DSPy empowers open-source AI development, allowing modular and compositional approaches for improving LLM-based AI systems over time.

---

📬 Get Involved

🚀 Contribute: Join our GitHub repo and submit PRs!
📢 Community: Engage in discussions on our Discord server.
📄 Documentation: Read the latest DSPy guides.

---

📌 Contact

📧 Email: saadsalmanakram1@gmail.com
🌐 GitHub: SaadSalmanAkram
💼 LinkedIn: Saad Salman Akram

---

🔥 Get Started with DSPy Today! 🔥