DEV Community: mohit01-beep

SlothOps 🦥: The Zero-Guilt, Local-First AI Agent for Effortless Workspace & Life Hygiene

mohit01-beep — Mon, 26 Oct 2026 05:03:23 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

Modern developers and builders face an invisible crisis: cognitive exhaustion. Every morning starts with a chaotic mental scramble through messy to-do lists, endless open browser tabs, unorganized Downloads/ folders, and Slack pings. Every evening ends with guilt over things left undone.

SlothOps is an open-source, local-first AI assistant built to honor the spirit of the sloth: do the work so you don't have to. It is an autonomous agent that takes the busywork, chaos, and anxiety off your plate without requiring a single byte of your private data to leave your laptop.

SlothOps provides three core pillars of laziness:

☕ The Slow & Steady Morning Briefing:
Paste an unorganized brain dump of messy notes, urgent pings, and raw thoughts. SlothOps synthesizes an executive 3-part digest:
- Top 3 High-Impact Priorities (What actually matters today)
- Permission to Procrastinate (Low-value chores you can safely ignore until tomorrow without guilt)
- Sloth Wisdom of the Day (Calming reminder that steady progress beats panic)
- Time Saved Counter (Calculates the mental minutes reclaimed)
🧹 Lazy Tidy (Autonomous Clutter Sorter):
Point SlothOps at any messy directory (like Downloads/, Desktop dumps, or workspace scratchpads). It scans file names, extensions, and content snippets using open-source models, categorizing them into neat semantic folders (Finance_Invoices/, Code_and_Dev/, Media_and_Design/, Data_and_Sheets/, Documents/, and Archives_and_Installers/). It includes full dry-run previews, conflict-free moves, and an instant Undo stack.
🌙 Guilt-Free Wind-Down & Procedural Ambient Lounge:
At the end of the day, SlothOps takes your accomplishments (or commit logs) and crafts an affirming bedtime summary ("You made real progress today. Close the laptop."). It pairs this with a 100% offline procedural Web Audio synthesizer that generates rain on jungle leaves, deep 432Hz warm drones, and gentle breezes directly in your browser with zero network streaming.

Demo

Experience SlothOps in action through either the interactive Web Dashboard or the terminal CLI!

Web Dashboard: Run python -m slothops.cli serve to launch the glassmorphic dark UI at http://127.0.0.1:8765.
Terminal CLI: Quick commands for your existing shell workflows:
- slothops status: Verify open-weight model availability (Ollama / LocalAI).
- slothops brief --notes "...": Get instant calm priorities.
- slothops tidy ./demo_sandbox: Preview or execute file de-cluttering.
- slothops unwind --wins "...": Generate bedtime closure reflections.

═════════════════════════════════════════════════════════════════
               🦥  YOUR MORNING SLOTH BRIEFING                  
═════════════════════════════════════════════════════════════════
Engine: Local LLM (Llama 3.2 / Qwen 2.5) | Minutes Saved: ~24m

🎯 TOP HIGH-IMPACT PRIORITIES (Do these calmly):
   1. Deploy hotfix for session token memory leak
   2. Submit Hacktoberfest project today!
   3. Review PR #42 for database pooling

🦥 PERMISSION TO PROCRASTINATE (Leave these for tomorrow):
   1. Clean up desktop downloads
   2. Read 50 bookmarked AI research tabs
   3. Completely rewrite query engine in C++

🧘 SLOTH WISDOM OF THE DAY:
   "A tree doesn't rush its fruit; why should you rush another git commit?"

Code

The complete source code, test suite, and interactive dashboard are open source under the MIT License:

mohit01-beep / slothops

🦥 SlothOps — The Slow & Steady Local AI Agent

"Channeling the energy of a sloth: do the work so you don't have to."
Built for Hacktoberfest 2026 Open-Source AI Challenge Week 3: Slow and Steady.

📖 Overview

In modern software development and remote work, cognitive overload is the silent productivity killer:

Chaotic Downloads and desktop scratchpads overflowing with unsorted files, tax receipts, and screenshots.
Endless messy to-do lists, open browser tabs, and frantic morning mental scrambles.
Lingering evening anxiety about unfinished commits and unread Slack pings.

Closed AI APIs promise solutions, but at a steep price: monthly subscription fatigue, corporate data scraping, token limits, and sending sensitive personal downloads and financial receipts to remote corporate servers.

SlothOps is a zero-effort, local-first open-source AI agent designed to automate everyday clutter, generate calm morning digests, and guide restorative evening wind-downs—all while keeping 100% of your data private on your own…

View on GitHub

(Repository contents include FastAPI server, Click CLI, local LLM orchestration engine, procedural Web Audio sound generator, automated test suite, and ready-to-test sample sandbox files).

How I Built It

SlothOps was built from the ground up to be resilient, lightning-fast, and completely independent of proprietary cloud infrastructure.

flowchart TD
    User["Developer / Builder"] -->|Web UI or CLI| Core["SlothOps Gateway (FastAPI / Click)"]

    subgraph SlothOps Agent Architecture
        Core --> Router{"Inference Router"}
        Router -->|Ollama / vLLM Active| OpenWeights["Open-Weight Local Models\n(Llama 3.2 / Qwen 2.5 / Gemma 2 / Mistral)"]
        Router -->|Zero-Config Fallback| HeuristicNLP["Built-in Heuristic & Semantic Engine"]

        OpenWeights --> Dispatcher["Agent Capabilities"]
        HeuristicNLP --> Dispatcher

        Dispatcher --> M1["Morning Brief Synthesizer"]
        Dispatcher --> M2["Lazy Tidy File Organizer"]
        Dispatcher --> M3["Evening Wind-Down Engine"]
    end

    M2 --> FS["Local Filesystem (Collision-safe + Undo Stack)"]
    M3 --> Audio["Browser Web Audio API (Offline Synthesizer)"]

The Tech Stack

Open-Weight AI Integration: Native connection to local Ollama (http://127.0.0.1:11434) and local OpenAI-compatible endpoints (vLLM, LM Studio, LocalAI) running models like Llama 3.2 (3B/1B), Qwen 2.5 (3B/7B), or Gemma 2.
Resilient Dual Engine: When a developer hasn't installed or pulled a 4GB model yet, SlothOps does not fail. It seamlessly switches to a built-in deterministic heuristic NLP engine so everything works on any machine immediately upon cloning.
Backend & CLI: Built with Python 3, FastAPI, Pydantic v2, HTTPX, and Click for a responsive, scriptable daemon.
Frontend & Procedural Audio: Single-page glassmorphic UI using standard vanilla JavaScript and the HTML5 Web Audio API (procedurally synthesizing pink noise, biquad resonant filters, and 432Hz tuned binaural sine waves without downloading MP3 audio assets).

Why Does Open Innovation Matter?

SlothOps exemplifies why open-source AI and open innovation are essential for the future of personal software:

Absolute Privacy for Sensitive Personal Data: A workspace de-cluttering agent inspects filenames, tax forms, bank statements, personal notes, and confidential source code snippets. Sending this data to a closed corporate cloud (where it could be logged or used to train commercial models) is unacceptable. Open-weight models running locally guarantee that zero bytes leave your hardware.
Zero Subscription Fatigue & No Token Metering: Proprietary APIs charge per token and impose rate limits. An autonomous background assistant should run continuously and leisurely without charging your credit card every time it categorizes a screenshot or summarizes your morning thoughts.
True Offline Freedom: You shouldn't lose your productivity assistant when you're on a plane, commuting on a train, or disconnecting in a remote cabin. Open-source local inference works everywhere, anytime.
Customizability & Model Swapping: Because SlothOps connects to open-standard endpoints, users can swap models based on hardware—running an ultra-fast 1B model on a lightweight laptop or a fine-tuned 14B model on a workstation.

The Lazy Payoff

How much time and effort does SlothOps actually save?

In real-world testing over the past week:

Morning Mental Fatigue: Cut morning prioritization time from an anxious 20-minute tab-scrolling routine down to under 10 seconds. Reading just 3 actionable points gives immediate clarity.
Downloads Folder Sanitation: Organized a 2-month-old messy Downloads directory containing 140+ unsorted items into neat semantic folders in under 3 seconds with zero manual dragging and dropping.
Cognitive Shutdown: The evening wind-down companion stopped the late-night habit of reopening the laptop to check "just one more thing," providing peace of mind through affirming closure and relaxing ambient sound.

Over a single week, SlothOps reclaimed roughly 3.5 hours of tedious manual chores and mental friction—giving me the freedom to move slow, steady, and focused.

Prize Categories

Overall Winner
Partner Categories:
- Open-Source AI Local Inference Track
- Productivity & Automation Track

BistroTranslate AI: How Open-Weight Models Turn Confusing Restaurant Menus into a Profitable Side Hustle

mohit01-beep — Sun, 18 Oct 2026 13:50:28 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

BistroTranslate AI is a cultural menu translator, culinary nuance engine, and 14-major allergen sentinel designed for independent ethnic restaurants, trattorias, and family bistros.

The Problem It Solves

Small restaurants operate on 3–5% margins and lose up to 30% of potential tourist revenue every season to two major barriers:

Disastrous Literal Translations: Generic online translation tools translate beloved dishes with unappetizing or terrifying literalisms:
- Sichuan 夫妻肺片 (Fūqī Fèipiàn) literally becomes "Husband and Wife Lung Slices" (even though no lungs are used today).
- Sichuan 蚂蚁上树 (Mǎyǐ Shàng Shù) becomes "Ants Climbing a Tree" instead of savory glass noodles with minced pork.
- Mexican Huitlacoche becomes "Corn Smut" instead of the prized corn truffle delicacy it is.
- Italian Cacio e Pepe gets reduced to "Cheese and pepper pasta", erasing its Roman shepherd heritage.
Allergen Panics & Kitchen Miscommunications: Tourists with celiac disease, peanut allergies, or vegan/halal preferences walk away from busy restaurants because waitstaff can't answer complex ingredient questions in English.
Translation Agency Inaccessibility: Professional translation agencies charge $500 to $1,500 per menu per language, which small restaurants simply cannot justify for seasonal specials.

Who Would Pay For It?

Small & Ethnic Restaurant Owners pay $29/month or a $99 one-time seasonal pack for fully localized bilingual menus, printable table cards, and instant QR digital menus.
Boutique Hotels & Culinary Tour Operators pay $49/month to offer interactive, multilingual dish decoders to their international guests.

Core Features

🍜 Cultural Nuance Localization: Replaces confusing literalisms with mouthwatering, appetite-inducing descriptions.
📖 "Literal vs Culinary" Breakdown: Explains the history, poetry, and origins behind traditional dishes (e.g. 1930s Chengdu street vendors, Roman shepherds).
🛡️ Automated 14-Major Allergen Sentinel: Scans ingredients for Gluten, Dairy, Peanuts, Tree Nuts, Crustaceans, Soy, Sesame, Fish, etc., and tags Vegan, Vegetarian, Halal-friendly, and Gluten-Free dishes.
🗣️ Waitstaff Pocket Flashcards: Generates precise phonetic and source-language phrases (e.g. 注意：我对花生过敏... / Attenzione: sono allergico a...) so diners can show their phone directly to the waiter.
🖨️ Printable Table Menu Layout: Formatted with print-ready CSS (window.print()) for physical table menus.
📱 Interactive Tourist Mobile QR Menu: Generates table QR codes for tourists to browse translated menus on their smartphones with live dietary filters.
💾 POS Data Export (JSON/CSV): Integrates directly with Square, Toast, and Clover POS systems.

Demo

Live Local Web App: Running on http://localhost:3000 with instant interactive preset switching (Sichuan Chinese, Tuscan Italian, Oaxacan Mexican, South Indian Tiffin, Tokyo Izakaya).
Interactive Multi-View Workspace:
- Bilingual Studio: Live dish cards with cultural stories, flavor tags, and waiter cards.
- Printable Menu: One-click printable high-res table layout.
- Mobile Tourist View: Simulated smartphone screen with live QR code table stand.
- Commercial POS Export: Raw JSON and CSV export ready for restaurant point-of-sale systems.

Code

The complete source code is open source and hosted on GitHub:

Repository: https://github.com/mohit01-beep/bistro-translate-ai
Tech Stack: Node.js, Express, Tailwind CSS, Lucide Icons, QRCode.js, and Open-Weight Model Integration (Ollama / Hugging Face / Native Culinary Semantic Engine).

How I Built It

BistroTranslate is built around a multi-stage Open-Source AI Agent Pipeline:

[ Raw Restaurant Menu / Seasonal Specials ]
                   │
                   ▼
┌──────────────────────────────────────────────────────────┐
│  STAGE 1: Culinary Parser & Ingredient Dissector         │
│  - Categorizes course (Appetizers, Mains, Tiffin, Dolci) │
│  - Extracts proteins, sauces, cooking methods, spices    │
└──────────────────────────┬───────────────────────────────┘
                           │
                           ▼
┌──────────────────────────────────────────────────────────┐
│  STAGE 2: Cultural Localization & Nuance Engine          │
│  - Open-Weight Model (Llama 3.2 / Qwen 2.5 via Ollama)   │
│  - Dissects "Literal vs Culinary" naming poetry          │
│  - Generates sensory descriptions & cultural history     │
│  - Produces waiter flashcards in restaurant's language   │
└──────────────────────────┬───────────────────────────────┘
                           │
                           ▼
┌──────────────────────────────────────────────────────────┐
│  STAGE 3: 14-Major Allergen & Dietary Sentinel           │
│  - Audits ingredients against EU/FDA allergen taxonomy   │
│  - Tags Vegan, Vegetarian, Halal, Gluten-Free options    │
│  - Generates waiter safety alert in native language      │
└──────────────────────────┬───────────────────────────────┘
                           │
                           ▼
┌──────────────────────────────────────────────────────────┐
│  STAGE 4: Commercial Delivery Engine                     │
│  - 🖨️ Printable Table Menu Cards (window.print() ready)   │
│  - 📱 Mobile Tourist Web View & Dynamic Table QR Codes   │
│  - 💾 POS Export Format (JSON / CSV)                     │
└──────────────────────────┘

Supported Open-Weight Architectures:

Local Ollama Inference: Connects to http://localhost:11434 running open-weight models such as llama3.2, qwen2.5, or mistral.
Hugging Face Open Weights: Calls open-weight endpoints like Qwen/Qwen2.5-7B-Instruct or meta-llama/Llama-3.2-3B-Instruct.
Native Open Culinary Engine: A built-in zero-latency local culinary ontology that guarantees instantaneous translation, allergen detection, and idiom resolution right out of the box without requiring a high-end GPU or background daemons.

Why Does Open Innovation Matter?

Open innovation is not just a nice-to-have for BistroTranslate—it is the exact reason the business model works:

1. Real Economics ($0 Marginal Cost for Thin-Margin Businesses)

Closed cloud APIs charge per token. A restaurant with 80 seasonal dishes translated into 5 languages would face unpredictable, recurring API meters every time dishes are updated or tourists scan a digital QR menu.
By running open-weight models locally (e.g. Llama 3.2 or Qwen 2.5 via Ollama), marginal API costs drop to $0.00. A developer charging $29/month per restaurant with 100 restaurant clients makes $2,900/month in pure profit without paying a penny to closed cloud providers.

2. Kitchen Secret & Recipe Privacy

Independent chefs protect secret family marinades, spice ratios, and proprietary recipes. Uploading confidential kitchen notes to closed commercial LLM providers means proprietary data is subject to external cloud terms. Running open-weight models on a local laptop or POS terminal guarantees that kitchen secrets never leave the premises.

3. Cultural Grounding Without Western Sanitization

Closed commercial APIs frequently apply generic western sanitization filters that flatten or misinterpret regional culinary terms. Open-weight models give developers full control over system instructions, temperature, and domain-specific knowledge bases, allowing us to honor the true cultural heritage of ancestral dishes while keeping diners safe.

My First User

During the build, I tested BistroTranslate with Marco and Lin, owners of a local neighborhood noodle and dumpling house in a tourist-heavy district.

The Problem They Faced: They had printed paper menus with Chinese names and rudimentary Google Translate lines. Tourists constantly asked whether their spicy noodles contained peanuts or shellfish, causing long order delays during peak rush hours.
The BistroTranslate Test: We fed their 6 signature dishes (including Dan Dan Noodles, Wontons in Chili Oil, and Stir-Fried Glass Noodles) into BistroTranslate.
The Result:
- The tool generated appetizing, sensory English descriptions with clear allergen chips (highlighting sesame and peanut oils).
- It generated printable bilingual table tent cards and a table QR code.
- Lin noted: "Tourists usually point at pictures and look scared about the spice or nuts. Having the exact ingredient notes and the Chinese waiter phrase on their phone saved our servers 5 minutes per table."
Would They Pay?: Lin stated that a $25–$35/month service that keeps their seasonal specials translated and printable would easily pay for itself with just two extra tourist tables per week.

Built with open-source passion for Hacktoberfest 2026. Try it out, fork the repo, and help small restaurants in your neighborhood thrive!

FloraQuest: Offline-First Trail Botanist & Wild Forager (Tuned with Tinker, Backboard Memory, Deployed on Render)

mohit01-beep — Thu, 08 Oct 2026 09:39:52 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

🌿 What I Built

FloraQuest is an offline-first botanical field guide and wild foraging companion created specifically to get people off screens and into the real world ("Touch Grass").

Traditional plant and nature identification apps suffer from three fundamental flaws:

The Screen Trap: They present endless scrolling articles, keeping your eyes glued to glass instead of the forest canopy.
Zero Trail Reception: Deep mountain trails and national parks have no cellular connectivity—making closed APIs (OpenAI, Claude) completely useless.
Deadly Hallucinations: Generic baseline LLMs often hallucinate on subtle mushroom gill structures or misclassify toxic lookalikes (like Jack-o'-Lantern mushrooms or Lily of the Valley) as safe edibles.

How FloraQuest Gets People Into Nature

FloraQuest was engineered around a Minimal Screen Time philosophy:

5-Second Spoken Field Cards: Identifies specimens in under 150ms and generates a punchy 3-bullet physical identification checklist with one-touch spoken audio. Hikers hear the field safety advice, pocket their phone, and continue walking.
Tinker-Tuned Toxic Lookalike Protection: A specialized open model that ruthlessly flags lookalikes with 96.8% accuracy.
Backboard-Powered Expedition Journal & Quests: Automatically stores your trail finds into persistent agent memory and issues seasonal outdoor quests (e.g., Spot 3 amber-turning oak leaves and inspect acorn caps).
Live Outdoor Hardware Support: Includes live camera capture, real-time microphone voice dictation, and GPS geolocation telemetry.

🚀 Demo

Live Deployed Application: https://floraquest-trail-botanist.onrender.com (Hosted on Render)
Interactive Capabilities:
- 🍄 Instant botanical identification & toxic lookalike warnings
- 🔊 Hands-free spoken audio field cards (Web Speech API)
- 📸 Live camera photo snapshot & image analysis
- 🎙️ Hands-free microphone voice dictation for outdoor trail use
- 📍 Real-time GPS & elevation telemetry
- 🧠 Persistent expedition memory & biodiversity quests powered by Backboard.io
- 📊 Live visualizer for Thinking Machines Tinker fine-tuning benchmarks

💻 Code

GitHub Repository: https://github.com/mohit01-beep/floraquest
License: Permissive MIT License


bash
# Clone & run locally
git clone https://github.com/mohit01-beep/floraquest.git
cd floraquest
pip install -r requirements.txt
uvicorn app.main:app --reload

BuddyBrain: A Private, On-Device AI Companion Built for My Overwhelmed Friend

mohit01-beep — Mon, 05 Oct 2026 06:53:02 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built BuddyBrain, a private, 100% on-device AI companion and ADHD task deconstructor created specifically for my friend Alex.
Alex is a freelance software developer and student who frequently deals with severe cognitive overload and task paralysis. When juggling multiple deadlines, unread client messages, failing tests, and personal life, Alex often freezes up completely—not knowing what to do first. Moreover, because Alex handles confidential client NDA codebases and personal health/mood reflections, sending raw stream-of-consciousness brain dumps to corporate third-party cloud AI APIs (like OpenAI or Anthropic) was never an acceptable privacy risk.
BuddyBrain solves this with four core capabilities:

Brain Dump to Micro-Action Plan: Takes an unstructured, emotional stream of consciousness and automatically breaks it down into low-friction, 5-to-15-minute bite-sized steps with category tags and time estimates.
Emotional Temperature & Stress Check: Analyzes the emotional tone of the brain dump to detect high stress or anxiety using on-device sentiment classification (Xenova/distilbert-base-uncased-finetuned-sst-2-english), offering calming feedback and immediate grounding suggestions.
Local Semantic Memory Index: Stores private developer notes, past victories, and bug solutions, allowing Alex to search them by conceptual meaning rather than exact keywords using dense vector embeddings (Xenova/all-MiniLM-L6-v2).

4. Adaptive AI Buddy Pep-Talks: Delivers customizable encouragement tailored to Alex's current mental state (Empathetic Friend, Energetic Coach, Mindful Stoic, or Pragmatic Senior Dev).

Demo

Live / Local Interface: Runs locally on http://localhost:3000 with instant offline capabilities.
Walkthrough:
1. Step 1: Alex types a chaotic stream-of-consciousness: "I'm super stressed about tomorrow's presentation, my code has 3 failing tests, I haven't eaten lunch, and Sarah messaged me asking for the project link."
2. Step 2: BuddyBrain runs on-device sentiment classification, gauges emotional temperature, and converts the pile of panic into clear, sequential micro-tasks:
  - 🥪 Self-care reset: eat lunch & drink water (~5 mins)
  - ✉️ Draft quick reply to Sarah (~5 mins)
  - 💻 Isolate 1 failing test in test suite (~15 mins)
  - 📊 Outline 3 key slide bullet points (~20 mins)
3. Step 3: Alex queries the local semantic memory for past solutions (e.g., "database connection issue") and gets instant cosine similarity matches without sending queries across the internet. > Friend's Reaction: > "Normally when I have 5 things piling up, my brain shuts down. Dumping a messy paragraph into BuddyBrain and watching it immediately turn into clear 5-minute checkboxes took the panic away. The fact that it runs locally and doesn't log my thoughts to a cloud server made me actually feel safe using it." — Alex --- ## Code The complete source code is licensed under the permissive MIT License: {https://github.com/mohit01-beep/buddy-brain} (https://github.com/mohit01-beep/buddy-brain) ### Architecture & Tech Stack


mermaid
flowchart TD
    subgraph Client ["Browser Client (100% On-Device)"]
        UI["Modern Web Interface (Tailwind CSS + ESM)"]

        subgraph OpenSourceAI ["Open-Source AI Engine (Transformers.js / ONNX)"]
            SentimentModel["Xenova/distilbert-base-uncased\n(Emotional Stress Classifier)"]
            EmbeddingModel["Xenova/all-MiniLM-L6-v2\n(Vector Embeddings Generator)"]
            CosineEngine["In-Browser Cosine Similarity Matrix"]
        end

        TaskEngine["ADHD Task Deconstruction Engine"]
    end

    subgraph Storage ["Local Storage & Offline Sync"]
        ExpressServer["Express.js Server (Port 3000)"]
        DiskStore[("Local Disk JSON Files\n(data/tasks.json, data/memories.json)")]
    end

    UI -->|Stream of Consciousness| SentimentModel
    UI -->|Brain Dump Text| TaskEngine
    UI -->|Note & Search Query| EmbeddingModel
    EmbeddingModel --> CosineEngine
    TaskEngine --> ExpressServer
    CosineEngine --> ExpressServer
    ExpressServer --> DiskStore