prompt-amplifier 0.1.1

Transform short prompts into detailed, structured instructions using context-aware retrieval

Prompt Amplifier 🔨

Transform short prompts into detailed, structured instructions using context-aware retrieval.

---

🎯 What is Prompt Amplifier?

Prompt Amplifier is a library for Prompt Amplification — the process of transforming short, ambiguous user intents into comprehensive, well-structured prompts that LLMs can execute effectively.

from prompt_amplifier import PromptForge

forge = PromptForge()
forge.load_documents("./company_docs/")

# Short, vague input
short_prompt = "How's the deal going?"

# Detailed, structured output
detailed = forge.expand(short_prompt)

Before (user input):

"How's the deal going?"

After (expanded prompt):

"Generate a Deal Health Assessment report with the following structure:

1. Executive Summary - Overall health status (Healthy/Warning/Critical)

2. Key Metrics Table

Metric	Current	Target	Status
Winscore	...	...	...
POC Progress	...	...	...

3. Risk Factors - List blocking issues from Success Plan

4. Recommended Actions - Prioritized next steps

Use data from: Deal Profile, Success Plan, Activity Log..."

---

✨ Features

• 📄 Multi-format Document Loading — PDF, DOCX, Excel, CSV, TXT, Markdown, HTML
• 🔢 Pluggable Embedders — TF-IDF, BM25, Sentence Transformers, OpenAI, Cohere, Google
• 💾 Vector Store Support — In-memory, ChromaDB, FAISS, Pinecone, Qdrant, Weaviate
• 🔍 Smart Retrieval — Vector search, hybrid (BM25 + Vector), reranking
• 🤖 LLM Backends — OpenAI, Anthropic, Google Gemini, Ollama (local)
• 📋 Domain Schemas — Define field structures for your domain
• 🔌 Extensible — Easy to add custom loaders, embedders, and vector stores

---

🚀 Quick Start

Installation

# Core library
pip install prompt-amplifier

# With common extras
pip install prompt-amplifier[loaders,embeddings-local,vectorstore-chroma]

# Everything
pip install prompt-amplifier[all]

Basic Usage

from prompt_amplifier import PromptForge

# Initialize with defaults (TF-IDF embedder, in-memory store)
forge = PromptForge()

# Load your documents
forge.load_documents("./docs/")

# Expand a short prompt
result = forge.expand("Give me a POC health check")

print(result.prompt)      # The expanded prompt
print(result.context)     # Retrieved context chunks
print(result.metadata)    # Stats and metadata

With Persistent Vector Store

from prompt_amplifier import PromptForge
from prompt_amplifier.vectorstores import ChromaStore
from prompt_amplifier.embedders import SentenceTransformerEmbedder

forge = PromptForge(
    embedder=SentenceTransformerEmbedder("all-MiniLM-L6-v2"),
    vectorstore=ChromaStore(
        collection_name="my_docs",
        persist_directory="./chroma_db"
    )
)

# First run: embeds and stores
forge.load_documents("./docs/")

# Subsequent runs: uses existing embeddings
result = forge.expand("Summarize the project status")

With Cloud Vector Store (Pinecone)

from prompt_amplifier import PromptForge
from prompt_amplifier.vectorstores import PineconeStore
from prompt_amplifier.embedders import OpenAIEmbedder

forge = PromptForge(
    embedder=OpenAIEmbedder(model="text-embedding-3-small"),
    vectorstore=PineconeStore(
        api_key="your-api-key",
        index_name="prompt-amplifier-prod"
    ),
    generator="gpt-4o"
)

---

📖 Documentation

• Getting Started
• API Reference
• Tutorials
• Research Paper

---

🧩 Architecture

┌─────────────────────────────────────────────────────────────┐
│                     Prompt Amplifier                         │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│   Documents    →   Chunker   →   Embedder   →   VectorStore │
│   (PDF, DOCX)      (split)       (encode)       (persist)   │
│                                                              │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│   User Query   →   Embedder  →   Retriever  →   Generator   │
│   "short"          (encode)      (search)       (expand)    │
│                                                              │
└─────────────────────────────────────────────────────────────┘

---

🔌 Supported Integrations

Document Loaders

Format	Loader
PDF	PDFLoader
Word	DocxLoader
Excel	ExcelLoader
CSV	CSVLoader
Text	TxtLoader
Markdown	MarkdownLoader
HTML	HTMLLoader
JSON	JSONLoader

Embedders

Provider	Class	Type
TF-IDF	TFIDFEmbedder	Free, Local
BM25	BM25Embedder	Free, Local
Sentence Transformers	SentenceTransformerEmbedder	Free, Local
OpenAI	OpenAIEmbedder	Paid API
Cohere	CohereEmbedder	Paid API
Google	GoogleEmbedder	Paid API
Voyage AI	VoyageEmbedder	Paid API

Vector Stores

Store	Class	Type
In-Memory	MemoryStore	Local
ChromaDB	ChromaStore	Local
FAISS	FAISSStore	Local
LanceDB	LanceDBStore	Local
Pinecone	PineconeStore	Cloud
Qdrant	QdrantStore	Local/Cloud
Weaviate	WeaviateStore	Cloud
pgvector	PGVectorStore	Self-host

LLM Generators

Provider	Class
OpenAI	OpenAIGenerator
Anthropic	AnthropicGenerator
Google Gemini	GoogleGenerator
Ollama	OllamaGenerator
HuggingFace	HuggingFaceGenerator

---

🧪 Research

Prompt Amplifier was developed as part of research into Prompt Amplification — systematically transforming short user intents into detailed, structured prompts.

Key contributions:

• Formalization of the prompt expansion problem
• Comparison of embedding strategies for prompt enhancement
• Evaluation metrics for prompt quality
• Benchmark datasets across multiple domains

📄 Paper: [Coming Soon]

---

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

# Clone the repo
git clone https://github.com/DeccanX/prompt-amplifier.git
cd prompt-amplifier

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linting
ruff check src/
black src/

---

📜 License

Apache 2.0 — See LICENSE for details.

---

🙏 Acknowledgments

Built with inspiration from:

• LangChain
• LlamaIndex
• ChromaDB

---

Made with ❤️ by Rajesh More for the AI community