DEV Community: Vimal Sahani

SlothPilot: The Autonomous Overnight Multi-Agent Command Center (Because Developers Deserve 8 Hours of Sleep)

Vimal Sahani — Wed, 21 Oct 2026 11:41:44 +0000

This is a submission for the Hacktoberfest 2026 Week 3 DEV Challenge: Slow and Steady

"Channel your inner sloth. You sleep 8 hours, SlothPilot works overnight."

In celebration of International Sloth Day (October 20) and the Hacktoberfest Week 3 Challenge ("Slow and Steady — Build something with open-source AI that rewards laziness"), I built SlothPilot.

🌐 Live Interactive Demo & Open-Source Code

Experience the Morning Command Center live in your browser:

👉 Live Streamlit App: https://slothpilot-xaqtgwwbpv2tfvscjhxakr.streamlit.app/
👉 GitHub Repository: https://github.com/VimalN2005/slothpilot

(You can simulate an overnight run right from the sidebar, inspect raw ingested sources, test the 1-click Approval Gate, generate an instant morning standup, listen to the 45-sec voice briefing, and inspect live OpenTelemetry trace spans!)

🥱 The Problem: The 9:00 AM "Digital Archaeology" Nightmare

Every software engineer knows the dreadful morning routine:
You sit at your desk at 9:00 AM with a cup of coffee. Instead of diving into creative, high-impact architecture, you spend your first 60 to 90 minutes performing stressful digital archaeology:

Dirty Git Branches: Which feature branches were left uncommitted before shutting the laptop last night?
Chaotic Chat Backlogs: What did teammates, managers, or clients message on WhatsApp, Discord, or Slack overnight? Did someone ping about an urgent payment webhook failure or an impending contract milestone?
Fragmented Notes & Todos: Scattered Markdown files, half-written bullet points in Obsidian/Notion, and priority tags (!high, p1, urgent) spread across multiple directories.
Calendar Anxiety: When is my first meeting today, and how much time do I actually have before I need to be camera-ready?
Standup Fatigue: Typing your morning standup update from memory while half-asleep.

By the time you finish triaging this chaos, your morning energy is drained. You've answered messages hastily, context-switched five times, and haven't written a single line of thoughtful code.

⚖️ The Contrast: The 9:00 AM Grinder vs The SlothPilot Developer

Feature	The 9:00 AM Corporate Grinder 😫	The SlothPilot Developer 🦥☕
Wakeup Ritual	6:30 AM cold panic checking Slack	9:45 AM (Enjoying +120m Sleep Buffer)
Git Archaeology	40 mins manually checking dirty branches	1-click approved auto-checkpoint
Client Triage	Scrambling 15 browser tabs & chat exports	Pre-drafted contextual replies ready
Daily Standup	15 mins typing updates from memory	1-click auto-generated Standup copied
Morning Briefing	Reading dense terminal diffs	🎙️ 45-sec calm audio briefing while brushing teeth
Privacy & Cost	\$50/mo cloud API + code sent to 3rd party	\$0.00 USD + 100% Local On-Device

💡 What I Built: SlothPilot 🦥

SlothPilot is an autonomous overnight multi-agent briefing system built with open-source AI and Python.

Before going to bed, you either let the automated scheduler trigger it at 03:00 AM or run a single command. While you sleep 8 uninterrupted hours, SlothPilot quietly inspects your local workspace, synthesizes overnight developments across 5 data sources, pre-drafts contextual replies and git actions, calculates your Sloth Sleep Buffer, and stages everything in a clean Morning Command Center with a strict 1-click Human Approval Gate.

When you wake up, you don't scramble through 10 browser tabs. You open your Morning Dashboard, listen to your audio briefing, copy your auto-standup, review your Top 3 Daily Focus Items, click Approve on pre-drafted client emails, and save 2+ hours every morning.

🏗️ How It Works: The Multi-Agent Pipeline

SlothPilot does not rely on a single massive prompt. Instead, it coordinates three specialized local agents orchestrated in a deterministic pipeline:

Local Data Sources (Git Repos, Notes, Chats, Todos, Calendar .ics)
        │
        ▼
   Ingestion Layer (app/ingestion)
   ├── Git Scanner (subprocess diffs & status)
   ├── Markdown Vault Parser (tasks & action items)
   ├── Chat Export Parser (WhatsApp/Discord urgent patterns)
   └── Backlog Parser (priority tags & deadlocks)
   └── Calendar Parser (.ics meeting timeline & Sleep Buffer)
        │
        ▼
   Multi-Agent Pipeline (app/agents)
   ├── 🧠 Context Aggregator Agent
   ├── ✍️ Action Proposer Agent
   └── 🛡️ Safety & Priority Reviewer
        │
        ▼
   Persistent State Store (SQLite)
   └── Full OpenTelemetry GenAI Span Traces
        │
        ▼
   Morning Command Center (Streamlit Dashboard)
   ├── 🎙️ 45-Sec Sloth Audio Briefing Player
   ├── 📋 1-Click Auto-Standup Generator (Slack / Discord)
   ├── 🛌 Sloth Sleep Buffer & Meeting Timeline
   ├── 🛡️ Human Approval Gate (1-click Approve / Dismiss)
   ├── 🔍 Ingested Workspace Inspector (5 Source Tabs)
   └── 📊 OpenTelemetry Observability (Latency & Zero-Cost Metrics)

1. The Context Aggregator Agent (`app/agents/aggregator.py`)

Scans raw local artifacts and extracts structured signals:

Uncommitted git diffs, dirty working trees, and recent commit messages.
High-priority chat patterns (BUG, URGENT, CONTRACT, PAYMENT, DEMO).
Open - [ ] markdown checkboxes and priority backlog todos (!high).
Upcoming morning meetings and schedule constraints from .ics exports.

2. The Action Proposer Agent (`app/agents/action_proposer.py`)

Rather than merely summarizing, this agent pre-drafts concrete solutions:

Client & Team Replies: Contextual, empathetic message drafts acknowledging the issue with specific next steps (e.g. SOW advance wire confirmation, payment webhook triage).
Git Workflows: Stash and branch commands (git add -A && git commit -m "checkpoint: overnight triage") to keep working trees clean.
Task Scheduling: Converts urgent messages directly into prioritized daily action cards.

3. The Safety & Approval Reviewer Agent (`app/agents/reviewer.py`)

Crucially, SlothPilot never executes side-effects unprompted. The reviewer agent:

Assigns safety ratings (SAFE, REQUIRES_REVIEW, DANGEROUS).
Calculates the Sloth Efficiency Score (measuring developer energy conserved).
Calculates the Sloth Sleep Buffer: If your first meeting isn't until 10:30 AM, SlothPilot informs you that you have +120 mins of extra guilt-free sleep!
Enforces the Human Approval Gate: all proposed actions remain staged until you explicitly click Approve or Dismiss.

👑 The Two Ultimate Laziness Features:

1. 📋 1-Click Auto-Standup Generator

Never rack your brain trying to remember what you committed yesterday or what meetings are scheduled today. SlothPilot generates an executive bulleted standup ready to paste directly into Slack or Discord:

🦥 Morning Standup Update (via SlothPilot):

🚀 Yesterday / Overnight:
- Triaged uncommitted git changes and overnight chat backlogs
- Automated test suite verification across workspace

🎯 Today's Focus:
- Review payment webhook queue blocker before 10 AM demo
- Approve and dispatch SOW contract reply to Client Dave
- Attend: Sprint Architecture & Payment Webhook Review (10:30 AM)

🛡️ Blockers:
- Needs approval: 2 pre-drafted client/team communications

2. 🎙️ 45-Second Voice Morning Briefing

For developers who refuse to look at screens before their first coffee, the dashboard includes a native browser voice player that reads your efficiency score, upcoming meeting time, and staged action counts aloud.

🛡️ The Human Approval Gate: Zero Autonomous Disasters

A major danger of autonomous AI agents is unprompted side-effects—an agent accidentally sending a premature email to a client or pushing broken code to production.

SlothPilot solves this with a Zero Unprompted Side-Effects Guarantee:

Every action card displays its category (COMMUNICATION, GIT_ACTION, TASK_RESOLUTION), safety rating, reasoning, and target destination.
With one click in the Streamlit UI, you can approve or dismiss each item.
Your morning workflow becomes: Review -> Approve -> Relax.

⏰ Set & Forget: Automated Overnight Scheduler Daemon

Prefer never to type a command before sleeping? SlothPilot includes an automated scheduler daemon (scripts/schedule_overnight.py):

python scripts/schedule_overnight.py --hour 3 --minute 0

SlothPilot quietly wakes up at 03:00 AM every night, runs the local ingestion and multi-agent pipeline, and stages your Morning Briefing before your alarm rings.

📊 OpenTelemetry Observability: Tracking Latency & Zero Costs

SlothPilot features built-in OpenTelemetry instrumentation (app/core/telemetry.py):

Every pipeline phase is tracked with W3C-compliant trace and span IDs (slothpilot.overnight_pipeline, slothpilot.ingestion, slothpilot.propose_actions, slothpilot.reviewer).
Real-time metric comparison: displays $0.00 USD inference cost vs the proprietary API tax you would have paid on closed cloud models.

🔓 Why Open Innovation & Open-Source AI Matter

SlothPilot operates on the most sensitive assets a developer possesses: uncommitted source code, private team chat exports, calendar meetings, and personal notes.

Sending this private data to cloud-hosted proprietary APIs poses significant privacy risks and unpredictable subscription fees. By utilizing open-weight models (via Ollama / Gemma / Llama):

100% Privacy & Zero Data Exfiltration: Your code, client conversations, and notes never leave your local machine.
$0.00 API Cost: Run overnight autonomous pipelines every single night without paying a dime in API bills.
Offline Autonomy: Works reliably even on flights or when your local network is down.

💻 Tech Stack

Language & Runtime: Python 3.10+
Data Validation & Schemas: Pydantic v2
Dashboard UI: Streamlit
Voice Synthesis: Native Web Speech API
State Persistence: SQLite
Observability: OpenTelemetry GenAI Spans
AI Core: Ollama / Open-Weight Models (Llama 3.1, Gemma 2)
CI/CD: GitHub Actions Matrix Testing (Python 3.10, 3.11, 3.12)
Test Suite: Pytest (13/13 unit tests passing in ~1.0s)

🦥 Final Thoughts: Embrace the Sloth Mindset

True engineering productivity isn't about waking up at 5:00 AM to answer 200 unread messages in a panic. It is about building resilient open-source tools that work diligently in the background so you can enjoy your sleep, wake up with clarity, and focus on what truly matters.

Happy Hacktoberfest 2026, and happy International Sloth Day! 🦥☕

CampusGemma AI: Turning Student Freelance Chaos into Profitable SOWs and Paid Mentorship with Google Gemma 2

Vimal Sahani — Tue, 13 Oct 2026 09:12:33 +0000

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

What I Built

As a final-year IT student at IIIT Bhopal, I watch classmates and fellow junior developers try to launch their first side hustle every single semester. Some want to build freelance React landing pages for local businesses, others want to offer 1-on-1 mock interview preparation for juniors, and others do freelance Python scripting.

Almost all of them hit the exact same brick wall in their first week:

The WhatsApp Brief Trap: Clients send chaotic, 15-sentence stream-of-consciousness WhatsApp messages with vague demands.
The Unpaid Scope-Creep Spiral: Because students don't know how to write binding legal Statements of Work (SOW), clients request "just one more tiny change" indefinitely.
The Proprietary API Tax: When students try adding AI to their side-hustle offerings, closed API token meters (OpenAI, Anthropic) eat away their micro-margins and instill billing anxiety.

To solve this, I built CampusGemma AI — a zero-marginal-cost micro-business studio powered by Google's open-weight Gemma 2 (gemma2-9b-it) and enterprise OpenTelemetry GenAI observability.

💼 The 5 Core Business Engines

Freelance SOW & Scope Creep Shield: Turns messy client briefs into professional Statements of Work with itemized deliverables, milestone USD quotes, and explicit anti-scope-creep boundaries ("Any revision exceeding 2 iterations billed at $25/hr").
Milestone Invoice & Settlement Studio: Auto-generates commercial milestone invoices with itemized deliverables, payment terms, and direct UPI / PayPal settlement instructions so student developers never get ghosted on payment.
Paid Mock-Interview Studio: Empowers senior students to monetize their placement experience by auto-generating structured technical interview rubrics, candidate rubrics, and answer keys.
Client Code & Security Auditor: A pre-delivery safety net that audits client code for exposed secrets, security leaks, and maintainability before handing over deliverables.
OpenTelemetry & Unit Economics Tracker: Real-time GenAI tracing that monitors latency, token usage, and confirms a 100% gross profit margin ($0.00 marginal inference cost).

Demo

Live Deployed App: https://campus-gemma-ai-bm7uyhdxvmwbfuy2ak5xkq.streamlit.app/
Zero-Config Offline / Demo Mode: Fully functional out of the box with realistic presets even if no API key is provided!

Key Walkthrough:

Generate & Export SOW in 1 Click: Pick one of the quick presets (e.g., Local Bakery Online Ordering or Student Club QR Ticketing), adjust your hourly rate, and click Generate Formal SOW & Quote. Within seconds, Gemma 2 produces a bulletproof client agreement with milestones and scope boundaries, ready to download via the "📥 Download Formal SOW Agreement (.md)" button.
Generate Commercial Milestone Invoices: Switch to the Invoice Studio to bill completed deliverables with custom UPI / PayPal payment instructions.
Monetized Mentorship Scorecards: Run 1-on-1 technical placement interviews or security code audits and instantly export structured candidate scorecards.
Inspect Live OpenTelemetry Tracing: Open the Observability & Economics tab to inspect active OpenTelemetry spans, latency breakdowns, and profit margin metrics confirming 100% gross margins.

Code

The complete codebase is open-source under the Apache 2.0 license:

GitHub Repository: https://github.com/VimalN2005/campus-gemma-ai
Automated CI/CD Test Suite: Passing 8/8 tests via GitHub Actions across Python 3.10, 3.11, and 3.12 covering proposal generation, invoice drafting, interview rubrics, security audits, and OpenTelemetry spans.

# Clone and run locally in under 60 seconds
git clone https://github.com/VimalN2005/campus-gemma-ai.git
cd campus-gemma-ai
pip install -r requirements.txt
streamlit run app.py

Core Architecture Snippet (`core/engine.py`):

def generate_freelance_proposal(client_brief: str, hourly_rate: int = 35) -> str:
    """Uses Google Gemma 2 to transform messy client chats into binding SOWs."""
    system_prompt = (
        "You are an elite freelance business advisor. Transform unstructured client briefs "
        "into structured, professional Statements of Work (SOW). Include: "
        "1. Executive Summary, 2. Itemized Deliverables, 3. Milestones & Pricing, "
        "and 4. Scope-Creep Protection clause."
    )
    user_prompt = f"Client Brief:\n{client_brief}\n\nTarget Rate: ${hourly_rate}/hr"
    return query_gemma_model(system_prompt, user_prompt)

How I Built It

1. The Open-Weight Core: Google Gemma 2 (`gemma2-9b-it`)

I selected Google's Gemma 2 9B Instruct model. Gemma 2 offers an extraordinary balance of reasoning capability, structured JSON adherence, and lightweight footprint. It follows complex contract formatting instructions with precision that rivals models 5x its size.

2. High-Speed Inference Engine

The application queries Gemma 2 via Groq's high-speed LPU infrastructure for near-instant (<1.2s) generation, with a built-in deterministic fallback engine for offline campus hackathons.

3. OpenTelemetry GenAI Instrumentation

As an active contributor to the CNCF OpenTelemetry Python GenAI ecosystem, I wired custom span tracing (core/telemetry.py) into every inference call. Every business action records:

gen_ai.system: gemma2
gen_ai.request.model: gemma2-9b-it
Latency and token throughput
Calculated marginal cost ($0.00)

Why Does Open Innovation Matter?

Why build this on open-weight models rather than a closed API like GPT-4 or Claude?

Zero Marginal Token Cost = Viable Micro-Hustles:

When a student charges $60 to build a student club website, every cent matters. If they rely on closed proprietary APIs for iterative client drafts, testing, and debugging, unexpected token bills can wipe out their weekend earnings. With open-weight models like Gemma 2, inference is free on local GPUs or zero-tier cloud endpoints. The business model works at unit volume 1.
Absolute Client Confidentiality & IP Security:

Freelancers frequently handle sensitive client materials — unreleased wireframes, confidential business logic, and proprietary codebases. Closed APIs retain rights to train or inspect payloads unless you pay for expensive enterprise tiers. Gemma 2 can be run 100% locally on a developer laptop with Ollama, ensuring zero client data ever touches external servers.
Freedom from Vendor Lock-In:

If a commercial AI API changes its terms of service, raises prices, or bans a use case, a small side hustle dies overnight. With Gemma 2, the weights belong to the open-source community forever.

My First User

For this challenge, I didn't want to stop at code. I tested CampusGemma AI with a real person: Aman, a 3rd-year Computer Science student at IIIT Bhopal who takes on freelance frontend projects.

The Real Problem:

Aman was dealing with a local client who kept sending rambling WhatsApp voice notes asking for "a clean portal, maybe add an admin dashboard too, and let's make it look like Airbnb". Aman had originally verbally agreed to build a simple 2-page website for \$100, but the client was now demanding a database, authentication, and multiple redesigns — without extra compensation.

The Test:

We pasted the client's messy WhatsApp conversation into CampusGemma AI's Freelance SOW Generator.

Within two seconds, Gemma 2 generated:

A crystal-clear 3-milestone breakdown.
An adjusted quote of \$160 reflecting the extra database work.
An explicit boundary clause: "Out of Scope: Additional design revisions beyond 2 rounds billed at \$25/hr."

The Result:

Aman exported the SOW as PDF and sent it back to the client. The client replied within 30 minutes: "This looks very professional, let's stick to Milestone 1 & 2 for now, here is the 50% deposit."

Aman earned his first protected freelance dollar with zero unpaid scope creep!

Prize Categories

I am officially entering CampusGemma AI into the following categories:

Featured Category: Best Use of Gemma (#gemma) CampusGemma AI is built directly around Google's open-weight Gemma 2 model (gemma2-9b-it), demonstrating how open-weight AI enables student freelancers and indie developers to operate profitable, privacy-first side hustles with zero API marginal costs.
Overall Challenge Winner

Built with ❤️ during Hacktoberfest 2026. Feedback and contributions are warmly welcome on GitHub!

Building TouchGrass AI: An Open-Source Nature & Outdoor Quest Companion to Rescue Developers from Burnout

Vimal Sahani — Tue, 06 Oct 2026 05:05:20 +0000

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

What I Built

As software engineers, we have all experienced that grueling state of mind: staring at an impenetrable concurrency deadlock, CSS layout bug, or broken build pipeline for 6 continuous hours until our eyes sting, our shoulders tense up, and our working memory collapses into mental exhaustion.

The developer community's timeless advice is simple: "Go touch grass."

For the Hacktoberfest 2026 Week 1 Challenge (Theme: "Touch Grass"), I built TouchGrass AI—an open-source intelligent companion designed to turn that meme into an accessible, science-backed outdoor reset protocol that gets engineers off the screen and back into the physical world.

Key Features:

🧘‍♂️ Custom Dopamine & Sunlight Quests: Developers input their screen time, current engineering task (debugging, architecture design, LeetCode grind), and symptoms (dry eyes, brain fog). TouchGrass AI synthesizes a personalized 10–15 minute outdoor quest with concrete sensory objectives (20-foot distance eye-strain relief, sunlight exposure, tactile foliage exploration) alongside a neuroscience rationale explaining how the exercise restores cognitive focus.
📷 Nature Vision & Flora Field Guide: When you step outside, snap or upload a photo of an interesting leaf, tree bark, or flower you encounter. The AI identifies the species, details its ecological role, and gives botanical trivia that awards gamified "Nature Points".
🎧 Guided Sensory Walking Stroll: A 4-phase audio-friendly strolling script tailored to dissolve bug loops from working memory so you can put on earbuds, pocket your phone, and walk hands-free.
🏆 Gamified Outdoor Streaks: Logs your outdoor days, tracks minutes in nature, and awards achievements like Sunlight Synthesizer and Botanical Explorer.

Demo

TouchGrass AI provides a clean, responsive interface built with Streamlit:

Instant Local UI: No complex setups—run streamlit run app.py and step away from the monitor in seconds.
Built-in Offline Engine: Includes a deterministic offline fallback mode so anyone can run it on a laptop in the park with zero internet connection or API keys!

Code

TouchGrass AI is 100% open-source under the MIT License:

VimalN2005 / touch-grass-ai

The Open-Source Nature & Outdoor Quest Companion for Software Developers (Hacktoberfest Week 1: Touch Grass).

🌿 TouchGrass AI

The Open-Source Nature & Outdoor Quest Companion for Software Developers.
Built for the Hacktoberfest 2026 Week 1 Challenge (Theme: "Touch Grass").

💡 Why TouchGrass AI?

As software engineers, we've all been there: staring at an impenetrable concurrency deadlock or CSS layout bug for 6 continuous hours until our eyes sting, our shoulders knot up, and our cognitive focus spirals into brain fog and frustration.

The internet's classic advice is simple: "Go touch grass."

TouchGrass AI transforms that meme into an intelligent, science-backed outdoor companion. Powered by open-source AI models (Meta Llama 3 open weights, local Ollama, and Hugging Face), it dynamically diagnoses your coding fatigue and synthesizes hyper-targeted, gamified outdoor micro-adventures to reset your neurotransmitters, relax your ciliary eye muscles, and restore your cognitive clarity.

✨ Key Features

1. 🎯 Custom Dopamine & Sunlight Reset Quests

Input your screen hours, current engineering task (debugging, PR reviews…

View on GitHub

GitHub Repository: https://github.com/VimalN2005/touch-grass-ai
Tech Stack: Python 3.10+, Streamlit, Pillow, Pytest, Open-weight LLMs (Meta Llama 3.1 / Ollama / Groq)

How I Built It

TouchGrass AI is engineered entirely around open-source AI and flexible local inference:

Open-Weight Model Inference: Powered by open weights (llama-3.1-8b-instant and llama-3.3-70b-versatile via Groq, or local Llama 3 via Ollama).
100% Local Execution with Ollama: Seamless native integration with local Ollama (http://localhost:11434), allowing complete offline privacy while sitting under a tree on a park bench.
Deterministic Nature Heuristics: Designed with an offline botanical engine so the app remains fully functional in deep hiking trails or signal dead-zones.

Why Does Open Innovation Matter?

Open innovation was non-negotiable for TouchGrass AI for three crucial reasons:

Complete Privacy on Health & Wellness Data: Mental health, burnout symptoms, and daily outdoor habits should never be harvested by proprietary advertising clouds. With open-weight models running locally via Ollama, zero bytes of user data ever leave your machine.
True Offline Portability on the Trail: Closed APIs require continuous cellular data. When you're hiking in deep woods or a national park with zero cell reception, proprietary APIs fail completely. Open weights allow intelligence to travel directly with you into nature.
Zero Operating Costs: Open-source AI eliminates recurring pay-per-token API fees. Any student, developer, or outdoor enthusiast can clone the repo and run it forever for free.

Prize Categories

Best Use of Open-Source AI
Community Wellness & Open Innovation

Next time your terminal throws an impossible stack trace, take a deep breath, step away from the keyboard, and let TouchGrass AI guide you outdoors! 🌿☀️

Building StudyBuddy AI for My College Friend: Turning Dense Notes into Interactive Quizzes with Open-Source AI

Vimal Sahani — Fri, 02 Oct 2026 03:42:08 +0000

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

💡 Who I Built This For ("Build for a Friend")

During exam weeks at my college (IIIT Bhopal), my friends and roommates often get overwhelmed by hundreds of slides of dense lecture notes—whether it is Operating Systems, Distributed Systems, or Machine Learning.

Reading passively through 50-page PDFs the night before an exam causes fatigue and poor retention. My friend Aman specifically told me: "I just wish there was a fast way to test myself on these key concepts without spending hours manually making flashcards."

That inspired me to build StudyBuddy AI.

🎓 What I Built

StudyBuddy AI is an open-source, interactive study companion built in Python. You simply paste your lecture notes or technical topic, and it immediately generates:

⚡ Executive Summary: 4–5 high-yield bullet takeaways.
🧠 Interactive Active-Recall Quiz: Custom multiple-choice questions with instant scoring, feedback, and in-depth explanations.
📌 Flash Cheat-Sheet: A clean glossary of core terms and definitions for quick revision right before the test.

🤖 Open-Source AI at the Core

To adhere to the core ethos of open-source and open-weight models, StudyBuddy AI is powered by Meta's Llama 3.1 (via fast inference APIs like Groq or Hugging Face):

It prompts the open-weight model to return strict, deterministic JSON schemas for summaries, quiz questions, and flash definitions.
Includes a zero-setup Demo / Offline Mode so anyone can test the full active-recall workflow locally without even requiring an API key.

💻 Tech Stack

Frontend & App: Python & Streamlit
AI Core: Open-weight LLMs (Meta Llama 3.1 8B Instruct)
Data Schemas: JSON-driven active recall structures

🔗 Code Repository

The code is 100% open-source and available on GitHub:

👉 GitHub Repository: https://github.com/VimalN2005/study-buddy-ai

🚀 What My Friend Thought

When I shared the prototype with Aman, he pasted his distributed systems notes and immediately took the 5-question generated quiz. The instant feedback on why an option was incorrect helped him grasp the CAP theorem and consensus protocols in minutes instead of hours.

Happy Hacktoberfest 2026! 🎉

DEV Community: Vimal Sahani

SlothPilot: The Autonomous Overnight Multi-Agent Command Center (Because Developers Deserve 8 Hours of Sleep)

🌐 Live Interactive Demo & Open-Source Code

🥱 The Problem: The 9:00 AM "Digital Archaeology" Nightmare

⚖️ The Contrast: The 9:00 AM Grinder vs The SlothPilot Developer

💡 What I Built: SlothPilot 🦥

🏗️ How It Works: The Multi-Agent Pipeline

1. The Context Aggregator Agent (app/agents/aggregator.py)

2. The Action Proposer Agent (app/agents/action_proposer.py)

3. The Safety & Approval Reviewer Agent (app/agents/reviewer.py)

👑 The Two Ultimate Laziness Features:

1. 📋 1-Click Auto-Standup Generator

2. 🎙️ 45-Second Voice Morning Briefing

🛡️ The Human Approval Gate: Zero Autonomous Disasters

⏰ Set & Forget: Automated Overnight Scheduler Daemon

📊 OpenTelemetry Observability: Tracking Latency & Zero Costs

🔓 Why Open Innovation & Open-Source AI Matter

💻 Tech Stack

🦥 Final Thoughts: Embrace the Sloth Mindset

CampusGemma AI: Turning Student Freelance Chaos into Profitable SOWs and Paid Mentorship with Google Gemma 2

What I Built

💼 The 5 Core Business Engines

Demo

Key Walkthrough:

Code

Core Architecture Snippet (core/engine.py):

How I Built It

1. The Open-Weight Core: Google Gemma 2 (gemma2-9b-it)

2. High-Speed Inference Engine

3. OpenTelemetry GenAI Instrumentation

Why Does Open Innovation Matter?

My First User

The Real Problem:

The Test:

The Result:

Prize Categories

Building TouchGrass AI: An Open-Source Nature & Outdoor Quest Companion to Rescue Developers from Burnout

What I Built

Key Features:

Demo

Code

VimalN2005 / touch-grass-ai

The Open-Source Nature & Outdoor Quest Companion for Software Developers (Hacktoberfest Week 1: Touch Grass).

🌿 TouchGrass AI

💡 Why TouchGrass AI?

✨ Key Features

1. 🎯 Custom Dopamine & Sunlight Reset Quests

How I Built It

Why Does Open Innovation Matter?

Prize Categories

Building StudyBuddy AI for My College Friend: Turning Dense Notes into Interactive Quizzes with Open-Source AI

💡 Who I Built This For ("Build for a Friend")

🎓 What I Built

🤖 Open-Source AI at the Core

💻 Tech Stack

🔗 Code Repository

🚀 What My Friend Thought

1. The Context Aggregator Agent (`app/agents/aggregator.py`)

2. The Action Proposer Agent (`app/agents/action_proposer.py`)

3. The Safety & Approval Reviewer Agent (`app/agents/reviewer.py`)

Core Architecture Snippet (`core/engine.py`):

1. The Open-Weight Core: Google Gemma 2 (`gemma2-9b-it`)