Forem: Krzysztof Nowicki

We just hit 100 GitHub stars ⭐ — and we’d love to meet some of you

Krzysztof Nowicki — Tue, 17 Mar 2026 18:12:33 +0000

We’ve just crossed 100 stars on GitHub for DocWire.

Not a huge number — but for a niche, modern C++ data processing SDK, it actually means a lot to us.

DocWire is built for:

extracting data from complex file formats
building offline / secure data pipelines
integrating with AI / LLM workflows (RAG, preprocessing, etc.)
all in modern C++20

We’re not a “viral” project.. (YET!)
Most of our users are solving very specific, often enterprise-level problems.

That’s why every single star matters.

A small request

If you starred the repo and you're here on Dev.to — I’d genuinely love to connect.

What are you working on?
What made you star the project?
Are you dealing with document/data extraction problems?

Even a short comment or message would be great.

What’s next

We’re actively developing:

new parsers (including EPUB)
better structured data extraction
easier "get up and go"
tighter integration with AI pipelines

Thanks again for the support — and curious to meet the people behind those stars ⭐ (here is the proof ! https://github.com/docwire/docwire )

First 2026 DocWire SDK Release: Modern C++20 XML Parsing, Safety Policies, and Parser Resilience

Krzysztof Nowicki — Thu, 22 Jan 2026 14:07:29 +0000

We’ve released a new version of DocWire SDK focused on core architecture, safety, and robustness, rather than user-facing features.

This update introduces a modern C++20 foundation for XML parsing, type-safe conversions, and configurable safety guarantees, alongside multiple improvements in parser resilience and memory management.

Highlights of this release

Modern C++20 XML parsing API
A new forward-only, single-pass XML reader based on C++20 ranges and views replaces the legacy XmlStream implementation.
Configurable safety policies
Developers can choose between strict checking (exceptions on violations) and relaxed, zero-overhead execution via checked, not_null, and enforce utilities.
Type-safe data conversion framework
New convert::try_to and convert::to APIs replace ad-hoc string conversions and support custom formats (e.g. date parsing).
Standardized date/time handling
All date and time logic now uses std::chrono::sys_seconds instead of struct tm.
Partial failure resilience
Parsers can continue processing even when some sub-items fail, while still detecting total failures.
New core utilities
Named parameters, non-null enforcement, ranged numeric types, and debug-only assertions.

In addition, the release includes parser-specific robustness fixes (HTML, PST, PDF), logging refinements, updated documentation, and expanded test coverage.

DocWire SDK in 2025 – Architecture, AI Pipelines, and Document Processing in Modern C++

Krzysztof Nowicki — Wed, 07 Jan 2026 11:37:51 +0000

In 2025, most of the work on DocWire SDK focused on architecture, correctness, and long-term maintainability, rather than surface-level features.

To document this work, we published a technical recap video summarizing the most important engineering changes introduced during the year.

DocWire SDK – 2025 Technical Summary

What changed in 2025

The video covers, among others:

migration from a std::variant-based data model to a polymorphic, message-driven pipeline architecture
exposing DocWire pipelines as HTTP/HTTPS microservices
integration of local, offline AI embeddings (multilingual E5 models)
expanded OpenAI support (GPT-4o, GPT-5, embeddings, transcription, TTS)
replacement of PoDoFo with Google PDFium
high-precision OCR and PDF positional metadata
image-aware PDF processing and OCR
modern HTML parsing and robust charset conversion
zero-cost logging, structured error diagnostics, and CI/CD modernization

The intent of this video is purely technical: architecture, APIs, performance, and engineering decisions.

Project links

GitHub: https://github.com/docwire/docwire
Website: https://www.docwire.io

If you're working on document processing, backend systems, or AI-assisted pipelines in modern C++, feedback and discussion are welcome.

Offline RAG in Modern C++: Secure Semantic Pipelines Without the Cloud

Krzysztof Nowicki — Wed, 10 Dec 2025 20:50:16 +0000

When you're dealing with confidential data — PII, medical records, trade secrets, or internal research — sending it to a third-party API for summarization or RAG preparation is a complete non-starter.

But that doesn’t mean you have to give up LLM power. With modern C++, you can build a universal, format-agnostic, fully offline data pipeline in just a few lines.

Below is how we (DocWire) generate embeddings for a PDF and a Word document, compare them for semantic similarity, and keep all data strictly on your machine — no cloud, no external API calls, no vendor lock-in.

- Define a secure offline pipeline

auto pipeline = content_type::detector{}
              | office_formats_parser{}
              | local_ai::embed(local_ai::embed::e5_passage_prefix);

This single chain handles:

format detection (PDF, DOCX, etc.)
file parsing
local embedding generation All offline.

- Process confidential documents locally

auto report_vec =
    std::filesystem::path("secret_plans.pdf") | pipeline;

auto policy_vec =
    std::filesystem::path("compliance_rules.docx") | pipeline;

No cloud calls. No data ever leaves your system.

- Compare semantic similarity

ensure(cosine_similarity(report_vec, policy_vec) > 0.85);

You now have a local-only RAG building block:

embeddings
comparisons
chunking
offline pipelines
zero dependency on OpenAI / Google / AWS Perfect for environments where data security is not optional.

Your turn : How do you handle secure, local-only RAG?
Different ecosystems approach this very differently. How would you design a cloud-free embedders + parser + similarity pipeline in: Python? Rust? Go? Java? C#? JavaScript?

Drop your snippet or architectural idea below

Modern C++ vs. The World: How Would Your Language Parse a Word Document?

Krzysztof Nowicki — Wed, 03 Dec 2025 17:07:27 +0000

Modern C++ isn’t dying. It’s eating file formats for breakfast.
Here’s what MS Word document parsing looks like today :

std::filesystem::path("data_processing_definition.doc")
    | content_type::detector{}
    | office_formats_parser{}
    | PlainTextExporter()
    | out_stream;

ensure(out_stream.str()) ==
    "Data processing refers to the activities performed on raw data...";

No COM, no Windows-only hacks, no XML archaeology - just a clean, composable pipeline in modern C++.
So now I’m genuinely curious: If this is what parsing looks like in modern C++, what does it look like in your favorite language?
Drop your snippet below.

DocWire SDK New Release: Turning Pipelines into Secure HTTP Services

Krzysztof Nowicki — Thu, 02 Oct 2025 12:30:51 +0000

DocWire SDK New Release: Turning Pipelines into Secure HTTP Services

We’re excited to announce a major new release of the DocWire SDK — and this one opens a completely new way to use DocWire in your applications.

At its core, DocWire has always been about high-performance document and data processing in C++. In this release, we’ve gone a step further:

We replaced the internal std::variant-based Tag system with a new polymorphic message_ptr architecture, simplifying the data flow and making it easier to extend.
And more importantly for many developers: we’ve introduced a built-in HTTP/HTTPS server that lets you expose DocWire pipelines as secure microservices.

Why the new HTTP server matters

Traditionally, to use DocWire you had to embed it directly into your C++ application. That works great — but many modern systems are moving toward service-oriented and cloud-native architectures.

With the new http::server class, you can now:

Expose any pipeline as a REST-like API: for example, mount a PDF-to-JSON parser at /parse/pdf.
Call DocWire from any language: Python, Java, Go, or even curl — if it can talk HTTP, it can use DocWire.
Scale horizontally: deploy pipelines behind a load balancer, containerize them with Docker/Kubernetes, and scale on demand.
Secure endpoints with TLS: essential for finance, government, and healthcare workloads.

This turns DocWire from a C++ library into a full-fledged data processing service framework.

A simple example

Let’s say you want to parse incoming PDF documents and expose the output as JSON through an API. With the new server:

docwire::http::server server;
server.add_pipeline("/parse/pdf", my_pdf_to_json_pipeline);
server.listen(8080, true); // true enables TLS

That’s it — now any client can POST a file to /parse/pdf and get structured JSON back.

What else is new in this release

Besides the HTTP server and the new polymorphic message system, we’ve included a number of improvements and fixes:

HTTP Post Content-Type: automatically set based on the input MIME type, making API integrations easier.
Enhanced vcpkg parsing: now supports more complex platform-specific feature definitions.
Reorganized element headers: document, mail, and AI element types are now separated for clarity.
Refactored XML options: dedicated XmlStream::no_blanks struct improves readability.
Unified transformer API: updated to the new message-based flow.
MSVC fixes: resolved min macro conflicts, added /bigobj for heavy templates.
macOS fixes: addressed RTTI visibility issues.
Dependencies: replaced curlpp with cpp-httplib and OpenSSL.
CI update: removed deprecated macos-13 runner.
Docs & tests: full migration to message_ptr, new end-to-end HTTP server integration tests.

Why this matters for developers

This release is not just about features — it’s about flexibility.

If you’re building monolithic C++ applications, you get a cleaner, more extensible architecture.
If you’re moving to microservices and distributed systems, DocWire now fits directly into that model.
If you need cross-language integration, the HTTP server makes DocWire accessible anywhere.

We believe this is a big step toward making DocWire SDK not just a C++ library, but a platform for data processing at scale.

Get started

The release is available now on GitHub: https://github.com/docwire/docwire/releases/tag/2025.09.25

As always, we welcome feedback, questions, and contributions from the community.

The DocWire Team -

DocWire SDK 2025.08.xx Released – GPT-5 Now Fully Integrated, New Default Model

Krzysztof Nowicki — Wed, 20 Aug 2025 14:10:46 +0000

The 2025.08.13 release brings a major milestone to DocWire SDK: full support for OpenAI's newly released GPT-5 family of models, including gpt_5, gpt_5_mini, and gpt_5_chat_latest.

To reflect this evolution, gpt_5 is now the default model for all OpenAI-related operations in DocWire — giving developers direct access to state-of-the-art AI with no additional configuration.

This version also brings minor but important code quality improvements and updated documentation.

Full release notes: https://github.com/docwire/docwire/releases/tag/2025.08.13

Features

1 · GPT-5 Model Family Support

Full integration of next-generation models:
- gpt_5, gpt_5_mini, gpt_5_nano, gpt_5_chat_latest
- Research-focused models: o3_deep_research, o4_mini_deep_research

2 · Default Model Upgraded

All OpenAI operations now default to gpt_5, replacing previous versions.

Improvements

Modern libxml2 Compatibility
- Deprecated xmlGetGlobalState() removed from XML parser initialization.
Better Error Handling
- Explicit exceptions now thrown for unknown or unsupported OpenAI models.
Code Quality Upgrades
- explicit added to AI-related single-argument constructors in AnalyzeData, ExtractEntities, and Summarize elements to avoid implicit conversions.

Documentation

Updated README
- Reflects new default model and the complete list of available OpenAI models
- CLI and code examples now demonstrate how to choose non-default models

Get Started

GitHub: https://github.com/docwire/docwire
Release: https://github.com/docwire/docwire/releases/tag/2025.08.13

We welcome feedback, examples, and issues as always.

— The DocWire Team

DocWire SDK 2025.08.05 Released – Local AI Embeddings, SentencePiece, Cosine Similarity

Krzysztof Nowicki — Thu, 07 Aug 2025 14:02:53 +0000

The 2025.08.05 release brings a major milestone to DocWire SDK: fully local, offline AI-powered text embeddings. With the integration of the multilingual-e5-small model, DocWire now supports multilingual vectorization for advanced NLP tasks—completely offline.

It also modernizes dependencies by switching from OpenNMT-Tokenizer to Google’s SentencePiece, and includes numerous build and CI improvements for better MSVC and Valgrind support.

Full release notes: https://github.com/docwire/docwire/releases/tag/2025.08.05

Highlights

1 · Local AI Embeddings

Introduces local_ai::embed for generating multilingual embeddings using multilingual-e5-small
Powers advanced use cases like semantic search, retrieval-augmented generation (RAG), and document clustering
CLI-ready via --local-ai-embed

2 · Cosine Similarity Utility

Built-in cosine similarity function for comparing document/query vectors

3 · Tokenizer API (SentencePiece)

Public local_ai::tokenizer based on Google’s SentencePiece
Supports encoding text into token IDs with T5Tokenizer and XLMRobertaTokenizer

Improvements

Unified model runner (local_ai::model_runner) now supports both encoder-only and sequence-to-sequence models
Advanced pooling and L2 normalization for E5-compatible output
New simplified constructor in model_chain_element with default model
CLI extended with support for embedding workflows

Refactors

Replaced OpenNMT-Tokenizer with a modern SentencePiece integration for improved maintainability and quality

Fixes

MSVC: AddressSanitizer (ASan) issues resolved using specific macro definitions
CI:
- Increased Valgrind timeouts
- Skipped heavy tests under Callgrind
- Abseil leak suppressions added for cleaner reports

Documentation & Tests

New end-to-end embedding example (README): document + queries + cosine similarity
Unit tests for local_ai::tokenizer
Embedding example is compiled and tested in CI

Get Started

GitHub: https://github.com/docwire/docwire
Release: https://github.com/docwire/docwire/releases/tag/2025.08.05
Sourceforge: https://sourceforge.net/projects/docwire/files/2025.08.05/

This update cements DocWire as a serious offline-ready NLP SDK for C++ developers building hybrid pipelines.

— The DocWire Team

DocWire SDK 2025.07.14 Released – OpenAI Embeddings, MIME Type Logic, HTTP Refactor

Krzysztof Nowicki — Wed, 16 Jul 2025 10:42:26 +0000

We’re back with a powerful new capability in DocWire SDK. Version 2025.07.14 adds support for OpenAI embeddings, modularizes the HTTP client, and improves MIME-based content handling across AI components.

Full release notes: https://github.com/docwire/docwire/releases/tag/2025.07.14

New Features

1 · OpenAI Embeddings Support

New openai::embed chain element enables developers to generate semantic embeddings from text using OpenAI’s latest models:
- text-embedding-3-small
- text-embedding-3-large
- text-embedding-ada-002
Includes CLI support (--openai-embed, --openai-embed-model) and full documentation with examples.

2 · MIME to Extension Utility

New content_type::by_file_extension::to_extension utility maps MIME types back to standard file extensions.

Improvements

More Robust AI Input Handling

AI components now detect input types using MIME (e.g., text/plain, image/png) instead of relying solely on file extensions. This improves accuracy in content classification and avoids misfires.
Smarter HTTP Uploads

The http::Post chain element can now infer file extensions from MIME types automatically, reducing friction in stream-based uploads.

Refactor

HTTP Client Modularization http::Post is now part of a dedicated docwire_http library, improving modularity and simplifying future expansion.

Fixes

Windows Build Fix Resolved vcpkg failure on Windows by explicitly linking libxml2.
README Fixed a formatting issue in the documentation.

A whisper of meaning, in numbers cast

New dimensions open, built to last

For search and reason, a powerful key

Unlocking knowledge for all to see

Try It Now

GitHub: https://github.com/docwire/docwire
Release: https://github.com/docwire/docwire/releases/tag/2025.07.14
Sourceforge https://sourceforge.net/projects/docwire/files/2025.07.14/

This release brings DocWire one step closer to being the go-to C++ SDK for hybrid AI-powered backends and document pipelines.

We welcome your feedback, issues, or questions.

— The DocWire Team

We are SourceForge RisingStars

Krzysztof Nowicki — Wed, 09 Jul 2025 07:01:33 +0000

DocWire SDK has received the Rising Star Award from SourceForge.

It’s awarded to open-source projects showing strong user traction and momentum — and with over 500,000 projects on SourceForge, that’s something we’re excited about.

If you haven’t heard of us:
DocWire is a modern C++ SDK for structured and unstructured data extraction, parsing, and transformation. It’s being used in production by consulting firms, AI tools, and secure environments — and we’re just getting started.

We are looking for:

Developers who care about modern, clean C++
Contributors who want to join an ambitious open-source project
Partners who need powerful data tooling or want to integrate with us
If that’s you, let’s talk.

Link to the project: https://github.com/docwire/docwire
SourceForge: https://sourceforge.net/projects/docwire/

Feedback, issues, and PRs are always welcome.

Thanks to SourceForge for the support and visibility.

-The DocWire Team-

DocWire SDK 2025.06.29 Released – New GPT-4o Support, Cleaner Builds, Smarter Prompts

Krzysztof Nowicki — Wed, 02 Jul 2025 22:43:15 +0000

A fresh DocWire SDK update is here. Version 2025.06.29 modernises our OpenAI integration, streamlines dependencies, and refines prompt engineering for more accurate AI-powered features.

Full release notes: https://github.com/docwire/docwire/releases/tag/2025.06.29

Key Features

1 · Expanded OpenAI Model Support

Added new models: gpt-4o, gpt-4o-mini, gpt-4.1, plus the new o3 family.
All AI-powered components now default to current-generation models.

2 · Granular Model Selection for Transcription and TTS

Transcription: choose among gpt-4o-transcribe, gpt-4o-mini-transcribe, or whisper-1.
TTS: new gpt-4o-mini-tts becomes the default for higher-quality voice synthesis.

3 · Dependency Modernisation

Replaced custom unzip vcpkg port with standard minizip, simplifying the build and improving maintainability.

Improvements

Enhanced Prompts – Classify and Find now use stronger system prompts for more precise, consistently formatted results.
Updated Default Model – General operations default to gpt-4o for better performance and cost efficiency.
Robust Example Tests – Documentation examples now use fuzzy string matching, avoiding false negatives from minor AI wording changes.

Refactor Highlights

API Clean-up – Deprecated OpenAI models (such as gpt-3.5-turbo and gpt-4-turbo-preview) have been removed.
Transcription Component – Refactored to support model selection, keeping the interface future-proof.

Get Started

GitHub repo – https://github.com/docwire/docwire
Latest release – https://github.com/docwire/docwire/releases/tag/2025.06.29
Sourceforce https://sourceforge.net/projects/docwire/files/2025.06.29/

We welcome feedback, issues, and PRs.

— The DocWire Team

DocWire SDK 2025.06.19 Released – Major OCR & PDF Layout Upgrades, Archive Refactor, CI Improvements

Krzysztof Nowicki — Wed, 25 Jun 2025 10:52:36 +0000

We’re back with a substantial update to DocWire SDK, the modern C++ library for structured document parsing, data extraction and secure, high-performance back-end workflows.

Version 2025.06.19 focuses on sharper OCR, more faithful PDF layout reconstruction and a brand-new archive module, alongside testing and CI upgrades.

Full release notes: https://github.com/docwire/docwire/releases/tag/2025.06.19

What’s New

1 · OCR Enhancements

Structured output with positional metadata – OCRParser now returns x, y, width, height plus line, paragraph and section grouping.
Configurable confidence filter (0–100) to ignore low-confidence words.

2 · Higher-Fidelity PDF Parsing

Refactored PDFParser to sort elements by position, yielding more accurate text flow and layout reconstruction.

3 · Modern Archive Handling

New docwire_archives library for modular, maintainable and faster archive processing.
Archive detection is now MIME-based.

4 · Expanded Format Support

Automatic detection for ASP and ASP.NET documents.

Developer-Centric Improvements

Plain-text exporter handles page breaks more clearly.
CI pipeline moves to windows-2025 runners; ASAN re-enabled on Windows.
Broader automated test coverage (OCR, HTTP, CLI).
Build fix on Windows via NOMINMAX flag to resolve windows.h / PDFium conflicts.
Spacing and line-break corrections in PDF and OCR outputs.

Documentation

API docs and module dependency notes are fully up to date.

OCR’s vision, sharp and newly bright

PDF layouts, now a clearer sight

Archives rebuilt, with structure firm and new

DocWire advances, steady, strong, and true

Try It Now

GitHub repo – https://github.com/docwire/docwire
Latest release – https://github.com/docwire/docwire/releases/tag/2025.06.19
Sourceforge - https://sourceforge.net/projects/docwire/files/2025.06.19/

We welcome feedback, issues and PRs.

Next up: deeper LLM integration and VCPKG support.

— The DocWire Team

Forem: Krzysztof Nowicki

We just hit 100 GitHub stars ⭐ — and we’d love to meet some of you

A small request

What’s next

First 2026 DocWire SDK Release: Modern C++20 XML Parsing, Safety Policies, and Parser Resilience

Highlights of this release

Links

DocWire SDK in 2025 – Architecture, AI Pipelines, and Document Processing in Modern C++

What changed in 2025

Project links

Offline RAG in Modern C++: Secure Semantic Pipelines Without the Cloud

Modern C++ vs. The World: How Would Your Language Parse a Word Document?

DocWire SDK New Release: Turning Pipelines into Secure HTTP Services

DocWire SDK New Release: Turning Pipelines into Secure HTTP Services

Why the new HTTP server matters

A simple example

What else is new in this release

Why this matters for developers

Get started

DocWire SDK 2025.08.xx Released – GPT-5 Now Fully Integrated, New Default Model

Features

1 · GPT-5 Model Family Support

2 · Default Model Upgraded

Improvements

Documentation

Get Started

DocWire SDK 2025.08.05 Released – Local AI Embeddings, SentencePiece, Cosine Similarity

Highlights

1 · Local AI Embeddings

2 · Cosine Similarity Utility

3 · Tokenizer API (SentencePiece)

Improvements

Refactors

Fixes

Documentation & Tests

Get Started

DocWire SDK 2025.07.14 Released – OpenAI Embeddings, MIME Type Logic, HTTP Refactor

New Features

1 · OpenAI Embeddings Support

2 · MIME to Extension Utility

Improvements

Refactor

Fixes

Try It Now

We are SourceForge RisingStars

DocWire SDK 2025.06.29 Released – New GPT-4o Support, Cleaner Builds, Smarter Prompts

Key Features

1 · Expanded OpenAI Model Support

2 · Granular Model Selection for Transcription and TTS

3 · Dependency Modernisation

Improvements

Refactor Highlights

Get Started

DocWire SDK 2025.06.19 Released – Major OCR & PDF Layout Upgrades, Archive Refactor, CI Improvements

What’s New

1 · OCR Enhancements

2 · Higher-Fidelity PDF Parsing

3 · Modern Archive Handling

4 · Expanded Format Support

Developer-Centric Improvements

Documentation

Try It Now