AI Enterprise SolutionsGPT & Claude

DeltaLens AI

DeltaLens is designed for office workers who regularly review digital documents, reports, and spreadsheets. Comparing multiple versions page by page is time-consuming, especially when the differences are small and easy to overlook. While existing tools can highlight changes, they may not explain them in the context of the document as a whole—leaving users unsure which changes matter. AI-generated summaries can help, but they may also miss or misinterpret important details. DeltaLens combines precise visual change detection with an optional plain-language summary, helping users quickly see exactly what changed and understand the bigger picture.

Open the demo Video GitHub

Preliminary score

47.5/ 100Earlyaverage of 2 judges

Value for the user13.0 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI13.5 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing9.0 / 20
Tests, examples of failures, and comparison with the current approach.
Feasibility8.0 / 15
Data requirements, running costs and a clear next step.
Originality4.0 / 10
A meaningful difference from common solutions.

What the judges wrote

GPT-6 AstraPreliminary50/100
Strengths
Your submission identifies a real document-review bottleneck and pairs visual differences with optional plain-language explanations. Your testing write-up usefully identifies font-conversion artifacts, shift-induced false differences, and AI-provider failures rather than claiming flawless performance. Your written submission also clearly explains the PDF/Office processing split and consent requirements for sending excerpts to AI providers.
To improve
Your deck describes code review, while your submission describes document comparison; align the product scope, setup instructions, and privacy claims, explicitly acknowledging server-side conversion and external AI calls. Show a reproducible document comparison through highlighting and AI explanation, demonstrating what AI contributes beyond deterministic diffing. Benchmark representative edits and formatting-only changes against manual review or an existing tool, reporting false positives, missed changes, and review time. Quantify conversion and AI running costs, and make correcting shift-induced false differences a concrete next milestone.
Claude Opus 5.5Preliminary45/100
Strengths
Your prototype covers a wide range of formats (PDF in-browser, plus DOCX/XLSX/PPTX via LibreOffice conversion) and adds a visual added/removed/modified highlighting layer, which is a real piece of engineering. Your testing write-up names concrete failures: Linux font substitution distorting DOCX-to-PDF output, line shifts flooding the diff with false changes, and unreliable AI providers. Those are honest, specific findings. Your privacy design is thoughtful: AI is opt-in, uses the user's own key, and sends only excerpts and metadata.
To improve
Your materials tell two different stories. The deck pitches semantic code diffs for pull requests, while the submission targets office workers comparing reports, and the README quick start points to a different repo ('docolens') with different run instructions. Pick one user and one workflow. For this enterprise track, measure a concrete metric: run a set of real document pairs, report the false-change rate before and after your line-shift fix, and compare review time against Word's Compare feature or a plain PDF diff. AI is currently an optional summary on top of a deterministic diff, so show what the LLM explanation adds and where it gets things wrong. Remove the self-assigned rubric scores and the 'Winning Submission' badge from your README.

The team

Yəhya Qurbanov
developer
Fuad Nasirli
developer
Zeynəb QurbanovaLead
developer

Disclosure

Required by the rules. Judges verify it against the repository.

Models: Claude 3.5 Haiku, GPT-4o-mini, Gemini 3.1 Pro, llama-3.3-70b-versatile, mai-code-1.1-flash
Data: No training datasets or proprietary data are used. Users provide the documents they want to compare: PDF, DOCX, XLSX, PPTX, ODT, ODS, ODP, CSV, or TXT files. PDF comparisons run in the browser. Office-format files are sent to the app’s server for conversion to PDF using LibreOffice before comparison. Optional AI explanations require the user’s own API key and consent; only comparison metadata and excerpts are sent to the selected AI provider. User documents are not used to train models.
Components: JavaScript, HTML/CSS, Node.js, Express, PDF.js, and a custom document comparison engine for page alignment, change detection, and visual highlighting. XLSX supports spreadsheet comparison; LibreOffice converts supported office documents to PDF. Optional AI explanations are available through provider integrations in the interface. The project also includes an embeddable JavaScript library and a command-line entry point.
Built during the event: Confirmed

Quality testing

We encountered three major challenges while testing DeltaLens. First, document conversion and formatting issues occurred because Linux did not properly support certain Microsoft fonts, causing DOCX-to-PDF conversion differences. As a result, users could not clearly identify the actual changes between documents, making it difficult to verify whether the program was working correctly. We addressed this by testing the documents on Zeynab Gurbanova’s Mac and Yahya Gurbanov’s Windows computers. Second, the diff algorithm produced misleading results. Even a small shift in a line or word could make large portions of a document appear different, overwhelming users with irrelevant changes and hiding the actual modifications. Finally, AI API reliability and cost created additional testing difficulties. Some providers charged for API access or did not work consistently, making it difficult to determine whether failures originated in our implementation or the provider’s service. We tested multiple providers and ultimately switched to Groq, using its Llama models throughout the project. These tests revealed that document comparison requires more than detecting textual differences: it must also handle formatting variations, line shifts, and inconsistent AI services without overwhelming users with false differences. bunu 500 karakter yap

Worth 20 of the 100 points.

Running it

DeltaLens — Setup Instructions Follow these steps to download, install, and run DeltaLens on your computer. No programming experience is required. 1. Download DeltaLens Open the DeltaLens GitHub repository. (https://github.com/ffunctia/deltalens) Click the green Code button. Select Download ZIP. Locate the downloaded ZIP file and extract it: Windows: Right-click the ZIP file and select Extract All. macOS: Double-click the ZIP file. Linux: Right-click the ZIP file and select Extract Here. Move the extracted project folder to a convenient location, such as your Desktop. 2. Install Node.js DeltaLens requires Node.js to run. Installing Node.js also provides npm, which is used to install the application's dependencies. Visit nodejs.org. Download the LTS (Long-Term Support) version. Open the installer and follow the instructions. The default settings are sufficient. Once installation is complete, close any open Terminal or Command Prompt windows and open a new one. To verify the installation, run these commands: node -v npm -v If both commands display version numbers, Node.js and npm are installed successfully. 3. Open a Terminal in the Project Folder Open the extracted DeltaLens folder. Make sure you are in the main project directory containing the package.json file. Windows: Open the folder in File Explorer, click the address bar, type cmd, and press Enter. Alternatively, right-click inside the folder and select Open in Terminal. macOS: Open Terminal, type cd (including the trailing space), drag the project folder into the Terminal window, and press Enter. Linux: Open a terminal in the project folder. Depending on your file manager, you may be able to right-click the folder and select Open in Terminal. 4. Install Dependencies and Start DeltaLens In the terminal, run these commands one at a time: npm install Wait for the installation to finish. This command downloads the packages required by the application. Then run: npm start Keep the terminal window open while using DeltaLens. Closing it will stop the application. 5. Open DeltaLens Open your web browser and visit: http://localhost:3000/deltalens You can now select two documents and start comparing them. Optional Requirements and Limitations PDF files: Comparing PDF documents does not require LibreOffice. Other document formats: To process DOCX, XLSX, PPTX, ODT, ODS, ODP, CSV, and TXT files, you need to install LibreOffice. Its soffice or libreoffice command must be available through your system's PATH. Without this configuration, conversion of these formats may not work. AI-powered explanations: AI explanations are optional. To use them, select an available AI provider in the application and enter your own API key. You do not need an API key for standard document comparison. Internet connection: DeltaLens loads some browser libraries from the internet, so an active internet connection is required.

Project ownership

The team keeps the rights to this project. Judges were given access for verification only.