<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Anirban Acharya</title>
    <description>The latest articles on DEV Community by Anirban Acharya (@ani04).</description>
    <link>https://dev.to/ani04</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4172917%2F7203eef7-0040-47ea-a39e-a3704683e9e0.png</url>
      <title>DEV Community: Anirban Acharya</title>
      <link>https://dev.to/ani04</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/ani04"/>
    <language>en</language>
    <item>
      <title>Ofrenda: A Local-First Memory Altar Powered by AI</title>
      <dc:creator>Anirban Acharya</dc:creator>
      <pubDate>Tue, 27 Oct 2026 06:06:53 +0000</pubDate>
      <link>https://dev.to/ani04/ofrenda-a-local-first-memory-altar-powered-by-ai-2olg</link>
      <guid>https://dev.to/ani04/ofrenda-a-local-first-memory-altar-powered-by-ai-2olg</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;What if you could preserve meaningful memories as candles and ask an AI to help you remember them — without sending your memories to a cloud AI service?&lt;/p&gt;

&lt;p&gt;That idea inspired Ofrenda, an experimental local-first memory altar built with HTML, JavaScript, and Ollama.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Ofrenda is inspired by Día de los Muertos, a tradition of remembering and honoring loved ones.&lt;/p&gt;

&lt;p&gt;The idea is simple: each memory becomes a candle. You can add memories, ask questions about what you've saved, and let a locally running AI model try to recall relevant details. When a memory is relevant to your question, its candle can be highlighted.&lt;/p&gt;

&lt;p&gt;The goal isn't to build another chatbot that answers everything. It's to explore an assistant that works with a small collection of personal memories and admits when it doesn't find something relevant.&lt;/p&gt;

&lt;p&gt;Features&lt;/p&gt;

&lt;p&gt;Memory candles: Add and save meaningful memories.&lt;/p&gt;

&lt;p&gt;Ask questions: Query the memories you've stored.&lt;/p&gt;

&lt;p&gt;Local AI: Generate responses using an Ollama model running on your computer.&lt;/p&gt;

&lt;p&gt;Relevant candle highlighting: Highlight memories matched to a question.&lt;/p&gt;

&lt;p&gt;A simple fallback: When no relevant memory is found, Ofrenda can respond with “I don't remember that.”&lt;/p&gt;

&lt;p&gt;Browser storage: Memories are saved in local storage.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/Anirban846" rel="noopener noreferrer"&gt;
        Anirban846
      &lt;/a&gt; / &lt;a href="https://github.com/Anirban846/ofrenda" rel="noopener noreferrer"&gt;
        ofrenda
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      A local-first memory altar for Día de los Muertos, powered by Ollama and open-weight AI.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;p&gt;# Ofrenda 🕯️&lt;/p&gt;
&lt;p&gt;**A local-first memory altar powered by Ollama and open-weight AI.**&lt;/p&gt;
&lt;p&gt;Ofrenda is a small experimental project inspired by Día de los Muertos. Add meaningful memories as candles, ask questions about them, and let a local AI model help you recall what you've saved.&lt;/p&gt;
&lt;p&gt;## Features&lt;/p&gt;
&lt;p&gt;- 🕯️ Add and save personal memories as candles&lt;/p&gt;
&lt;p&gt;- 💬 Ask questions about saved memories&lt;/p&gt;
&lt;p&gt;- 🧠 Uses Ollama to run an AI model locally&lt;/p&gt;
&lt;p&gt;- 🔒 Keeps memories in browser local storage and sends selected memories to the local Ollama service&lt;/p&gt;
&lt;p&gt;- ✨ Highlights candles relevant to a question&lt;/p&gt;
&lt;p&gt;- 🌱 Responds when it cannot find a relevant saved memory&lt;/p&gt;
&lt;p&gt;## Requirements&lt;/p&gt;
&lt;p&gt;- Python 3&lt;/p&gt;
&lt;p&gt;- [Ollama](&lt;a href="https://ollama.com/" rel="nofollow noopener noreferrer"&gt;https://ollama.com/&lt;/a&gt;)&lt;/p&gt;
&lt;p&gt;- A supported Ollama model, such as &lt;code&gt;llama3.2:3b&lt;/code&gt;&lt;/p&gt;
&lt;p&gt;## Run locally&lt;/p&gt;
&lt;p&gt;1. Install Ollama and download the model:&lt;/p&gt;
&lt;p&gt;   ```powershell&lt;/p&gt;
&lt;p&gt;   ollama pull llama3.2:3b&lt;/p&gt;
&lt;p&gt;   ```&lt;/p&gt;
&lt;p&gt;2. Open PowerShell in the project directory and…&lt;/p&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/Anirban846/ofrenda" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Ofrenda combines a simple browser interface with a locally running language model.&lt;/p&gt;

&lt;p&gt;You add memories, which are stored in your browser.&lt;/p&gt;

&lt;p&gt;When you ask a question, lightweight text matching identifies potentially relevant memories.&lt;/p&gt;

&lt;p&gt;Ofrenda sends the selected memories and your question to the local Ollama service.&lt;/p&gt;

&lt;p&gt;The model generates an answer based on the supplied context.&lt;/p&gt;

&lt;p&gt;A basic text-grounding check evaluates the answer, and relevant candles are highlighted.&lt;/p&gt;

&lt;p&gt;This approach keeps the project small and easy to experiment with. It also has limitations: text matching can miss relevant memories, and a lightweight grounding check cannot guarantee that every answer is correct.&lt;/p&gt;

&lt;p&gt;🔒 Why Local-First AI?&lt;/p&gt;

&lt;p&gt;Personal memories can be sensitive. I wanted to explore an approach that doesn't depend on a hosted AI API.&lt;/p&gt;

&lt;p&gt;Ofrenda stores memories in browser local storage and sends selected memories to the local Ollama service for answering. It doesn't require a cloud AI API.&lt;/p&gt;

&lt;p&gt;This is a local-use design, not a guarantee of complete security. Browser storage isn't encrypted by default, and the computer running the application remains part of the privacy boundary.&lt;/p&gt;

&lt;p&gt;🛠️ Tech Stack&lt;/p&gt;

&lt;p&gt;HTML&lt;/p&gt;

&lt;p&gt;CSS&lt;/p&gt;

&lt;p&gt;JavaScript&lt;/p&gt;

&lt;p&gt;Ollama&lt;/p&gt;

&lt;p&gt;llama3.2:3b (local language model)&lt;/p&gt;

&lt;p&gt;🚀 Run It Locally&lt;/p&gt;

&lt;p&gt;Prerequisites: Python 3, Ollama, and a compatible local model.&lt;/p&gt;

&lt;p&gt;First, download the model:&lt;/p&gt;

&lt;p&gt;ollama pull llama3.2:3b&lt;/p&gt;

&lt;p&gt;Clone the repository:&lt;/p&gt;

&lt;p&gt;git clone &lt;a href="https://github.com/Anirban846/ofrenda.git" rel="noopener noreferrer"&gt;https://github.com/Anirban846/ofrenda.git&lt;/a&gt;&lt;br&gt;
cd ofrenda&lt;/p&gt;

&lt;p&gt;Start a local web server:&lt;/p&gt;

&lt;p&gt;python -m http.server 8000 --bind 127.0.0.1&lt;/p&gt;

&lt;p&gt;Open the app in your browser:&lt;/p&gt;

&lt;p&gt;&lt;a href="http://127.0.0.1:8000/ofrenda.html" rel="noopener noreferrer"&gt;http://127.0.0.1:8000/ofrenda.html&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Make sure Ollama is running and the model name in the app matches your installed model. If your browser blocks requests to Ollama, you may need to configure Ollama's allowed origins for the local address and restart Ollama.&lt;/p&gt;

&lt;p&gt;🧪 What I Tested&lt;/p&gt;

&lt;p&gt;During development, I tested two important cases:&lt;/p&gt;

&lt;p&gt;A known memory: When I saved a memory about adding salt until water tastes like the sea, Ofrenda returned the saved instruction.&lt;/p&gt;

&lt;p&gt;An unknown memory: When I asked about a movie that wasn't in the saved memories, Ofrenda returned its “I don't remember that” fallback.&lt;/p&gt;

&lt;p&gt;These tests helped me improve the app's handling of missing information. They don't prove that hallucinations are impossible, but they demonstrate the behavior I wanted to explore.&lt;/p&gt;

&lt;p&gt;💡 What I Learned&lt;/p&gt;

&lt;p&gt;Building Ofrenda reinforced an important lesson: telling an AI model not to hallucinate isn't enough.&lt;/p&gt;

&lt;p&gt;A more reliable design needs to consider which memories are relevant, what context the model receives, and whether its response is supported by that context. Even simple text-based checks can introduce false positives and false negatives, so they need to be treated as safeguards rather than guarantees.&lt;/p&gt;

&lt;p&gt;I also explored how a local model can support an AI experience without requiring a hosted inference API.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open innovation gives developers room to experiment with ideas that might otherwise remain just concepts. Using accessible tools and open-weight models made it possible to build and test this small project locally.&lt;/p&gt;

&lt;p&gt;Ofrenda is an experiment, and there is plenty of room to improve its memory retrieval, reliability, accessibility, and visual design.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Spirit of It
&lt;/h2&gt;

&lt;p&gt;Ofrenda started with a simple thought: technology should help us remember, not just generate answers.&lt;/p&gt;

&lt;p&gt;Inspired by Día de los Muertos, this project treats memories as something worth preserving, with each candle representing a moment we don't want to lose. The AI isn't meant to invent a life story or pretend to know everything. It's meant to help us revisit what we've chosen to remember — and acknowledge when something isn't there.&lt;/p&gt;

&lt;p&gt;That idea shaped one of the most important parts of the project: “I don't remember that” should be a perfectly acceptable answer.&lt;/p&gt;

&lt;p&gt;I also wanted to explore a more personal relationship with AI, one where memories can stay on our own computers instead of being sent to a hosted AI service. Ofrenda is a small experiment, and it isn't a perfect memory system, but it represents the kind of technology I want to explore: more thoughtful about privacy, more honest about uncertainty, and built around people rather than just prompts.&lt;/p&gt;

&lt;p&gt;Not every memory needs to be generated. Some simply need a candle. 🕯️&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>ai</category>
      <category>javascript</category>
    </item>
    <item>
      <title>sloth-tidy</title>
      <dc:creator>Anirban Acharya</dc:creator>
      <pubDate>Sun, 25 Oct 2026 06:15:12 +0000</pubDate>
      <link>https://dev.to/ani04/sloth-tidy-1m3e</link>
      <guid>https://dev.to/ani04/sloth-tidy-1m3e</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week3-2026-10-19"&gt;Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;sloth-tidy is a tiny command-line tool that cleans up a messy folder so you don't have to. It reads each file's name (and the start of its text, if it's a text file), asks a local AI model what it is, then sorts it into a folder like &lt;code&gt;Documents&lt;/code&gt; or &lt;code&gt;Archives&lt;/code&gt; and gives it a readable name. &lt;code&gt;Scan 001.txt&lt;/code&gt; becomes &lt;code&gt;Documents/invoice-march-electricity.txt&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It's built for the "I'll sort my Downloads folder this weekend" crowd, which is to say, me.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/Anirban846/sloth-tidy" rel="noopener noreferrer"&gt;https://github.com/Anirban846/sloth-tidy&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Model: any open-weight model served locally by &lt;a href="https://ollama.com" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt;. I defaulted to &lt;code&gt;llama3.2:3b&lt;/code&gt; because it's small enough for a laptop.&lt;/li&gt;
&lt;li&gt;Approach: one file per request, with the model forced to return JSON (&lt;code&gt;format: json&lt;/code&gt;). I validate the reply: unknown categories become &lt;code&gt;Other&lt;/code&gt; and names are slugified, so a bad answer can't produce a bad path.&lt;/li&gt;
&lt;li&gt;Safety, because lazy people don't check: dry run by default, a collision-proof renamer that never overwrites, and an undo log so &lt;code&gt;--undo&lt;/code&gt; puts everything back.&lt;/li&gt;
&lt;li&gt;Dependencies: none. Python's standard library talks to Ollama's local HTTP API.&lt;/li&gt;
&lt;li&gt;Fallback: if the model isn't running, it sorts by file extension instead of crashing.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;A file tidier reads your files. Receipts, scans, half-finished documents. Sending those to a closed API just to rename them felt wrong, so with an open-weight model the whole thing runs on my own machine with no account, no key, and no per-call cost. Set it up once and it's free forever.&lt;/p&gt;

&lt;p&gt;It also means I can swap models with one flag (&lt;code&gt;--model&lt;/code&gt;) as better small ones appear, and fix the behavior with a prompt edit instead of waiting on a vendor.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>LocalInvoiceAI: Private, AI-Powered Invoice Extraction on Your Own Machine</title>
      <dc:creator>Anirban Acharya</dc:creator>
      <pubDate>Sun, 18 Oct 2026 16:39:25 +0000</pubDate>
      <link>https://dev.to/ani04/localinvoiceai-private-ai-powered-invoice-extraction-on-your-own-machine-5e5i</link>
      <guid>https://dev.to/ani04/localinvoiceai-private-ai-powered-invoice-extraction-on-your-own-machine-5e5i</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;LocalInvoiceAI is a locally running AI-powered invoice extraction and management app designed to help freelancers, independent professionals, and small businesses reduce the manual work involved in processing invoices.&lt;br&gt;
Invoices contain important information—invoice numbers, dates, vendor details, subtotals, taxes, and totals. Manually entering this information into spreadsheets or accounting systems can be repetitive and time-consuming.&lt;br&gt;
LocalInvoiceAI lets users upload invoice documents and extract structured information using a local AI model. Users can review and correct the extracted fields before saving the invoice.&lt;br&gt;
Key features:&lt;br&gt;
📄 Invoice extraction: Extract invoice details using a locally running AI model.&lt;br&gt;
🔎 PDF and image support: Read text from PDFs and use OCR to process scanned documents and invoice images.&lt;br&gt;
✏️ Human-in-the-loop review: Review and edit extracted information before saving.&lt;br&gt;
✅ Financial validation: Check basic consistency between invoice amounts and totals.&lt;br&gt;
🗄️ Local database: Store invoice records in SQLite.&lt;br&gt;
📊 Dashboard: View invoice records and useful summary information.&lt;br&gt;
🔍 Search and export: Find saved invoices and export records to CSV.&lt;br&gt;
The goal is simple: make invoice processing more convenient without making a cloud AI service a mandatory part of the extraction workflow.&lt;br&gt;
Who could benefit? Freelancers, small business owners, independent contractors, and anyone who regularly handles invoices.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;GitHub repository: &lt;a href="https://github.com/Anirban846/LocalInvoiceAI" rel="noopener noreferrer"&gt;https://github.com/Anirban846/LocalInvoiceAI&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;LocalInvoiceAI combines traditional document processing with local AI inference.&lt;br&gt;
Tech stack&lt;br&gt;
Python — application logic and data processing.&lt;br&gt;
Streamlit — interactive web interface.&lt;br&gt;
Ollama — local model runtime.&lt;br&gt;
Qwen2.5 3B — open-weight language model used for structured information extraction.&lt;br&gt;
Tesseract OCR — text recognition for scanned invoice images.&lt;br&gt;
pypdf — text extraction from PDF documents.&lt;br&gt;
Pillow — image processing.&lt;br&gt;
SQLite — persistent local invoice storage.&lt;/p&gt;

&lt;p&gt;How the workflow works&lt;br&gt;
The user uploads an invoice.&lt;br&gt;
The application extracts text directly from PDFs or uses OCR when processing images.&lt;br&gt;
The extracted text is passed to the locally running Qwen2.5 3B model to identify relevant invoice fields.&lt;br&gt;
The application checks basic financial consistency.&lt;br&gt;
The user reviews and corrects the extracted values.&lt;br&gt;
The verified invoice can be saved to SQLite and included in searchable records and CSV exports.&lt;/p&gt;

&lt;p&gt;One important design decision was to keep the user involved in the process. AI extraction is not guaranteed to be correct, especially when documents have poor image quality, unusual layouts, or ambiguous financial fields. Reviewing the original document before saving remains important.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;Open innovation made it possible to experiment with a useful AI workflow without making a paid, hosted language-model API a core dependency.&lt;/p&gt;

&lt;p&gt;Using an open-weight model through Ollama gives me control over where inference runs and makes it possible to develop and test the application on my own computer.&lt;/p&gt;

&lt;p&gt;It also makes experimentation more accessible. I can inspect the extracted results, adjust the extraction logic, improve validation, and try different models without redesigning the entire application around a single hosted AI provider.&lt;/p&gt;

&lt;p&gt;Running inference locally can help reduce the amount of invoice data sent to external AI services. However, local inference alone does not guarantee complete privacy: document storage, logs, backups, and other application components must also be handled carefully.&lt;/p&gt;

&lt;p&gt;For me, this project demonstrates how open AI tools can turn a practical, repetitive business task into an application that people can run and adapt themselves.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
  </channel>
</rss>
