DEV Community: 3078 Richa Kumari

ColdStudio

3078 Richa Kumari — Sun, 18 Oct 2026 16:24:39 +0000

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

What I Built

I built ColdPulse Studio, a local-first B2B cold outreach engine and prospect intelligence tool designed for solo founders, freelancers, sales development reps (SDRs), and agency operators building outbound side hustles

Who would pay for it and what problem does it solve?

Outbound lead generation is one of the highest-leverage skills for anyone starting a consulting, agency, or B2B side hustle, but the existing tooling ecosystem works against beginners:

Expensive SaaS Tool Tax: Mainstream AI SDR tools charge anywhere from \$50 to \$200+ per seat per month, draining early profit margins before a solo founder closes their first client.
Generic, Identifiable AI Spam: Most platforms use basic prompt templates over closed models that generate recognizable robotic clichés like "I hope this email finds you well"
Data Privacy & Third-Party Retention: Routing client prospects, custom offers, and proprietary company notes through remote cloud APIs creates vendor lock-in and corporate data-privacy concerns.

ColdPulse Studio solves this by putting an autonomous research and copywriting pipeline directly onto your laptop. You enter your target prospect's website and your value proposition. The engine automatically extracts and sanitizes live website text, feeds the context into a local open-weight model via Ollama, and crafts structured, hyper-personalized outreach in seconds—with **₹0 in recurring API fees, zero data retention, and 100% offline control

Key Workflow Highlights

Live Content Scraper: Strips away CSS, scripts, and navigation clutter from the target URL to extract real business copy and meta descriptions.
Interactive Email Client Canvas: Visualizes the rendered draft in a simulated macOS/webmail container with dynamic subject line choices and icebreaker hook inspection.
One-Click Dispatch & CRM Sync: Open drafts directly in your default email application via pre-filled mailto: links, or export leads and subject lines into structured .csv files.

Code

The full code is open source and ready to clone:

Richs03 / Week-Challenge-2

Week-Challenge-2

⚡ ColdPulse Studio

Precision B2B cold outreach platform powered entirely by local open-weight LLMs. Zero API costs, zero data retention, and fully autonomous website intelligence.

ColdPulse Studio is designed for solo founders, freelancers, agency operators, and consultants running lead generation side hustles. Instead of paying hefty monthly subscriptions for AI SDR tools, ColdPulse scrapes your prospect's public footprint and drafts bespoke, high-converting outreach locally using Ollama.

✨ Features

🌐 Live Context Scraping: Extracts clean text, headers, and meta descriptions from target company domains, filtering out navigational boilerplate.
🔒 100% Local & Cost-Free: Runs on local open-weight models (qwen2.5:3b, llama3.2:1b) via Ollama. No OpenAI tokens or external API bills.
✉️ Simulated Mailbox Canvas: Visualizes the rendered draft in an authentic client UI with selectable subject lines and icebreaker inspection.
🎯 Strategy Customization: Tune tone, target length, and campaign objectives (e.g., Academic Partnership, Direct B2B, Discovery).
🚀 One-Click…

View on GitHub

How I Built It

ColdPulse Studio is engineered with a strict separation between user interface, text extraction, and local LLM orchestration:

Local Inference Engine (Ollama + qwen2.5:3b / llama3.2:1b): The core intelligence runs entirely on-device using open-weight models loaded into Ollama. Rather than receiving unstructured markdown, the model is bound by a strict JSON contract (format="json"). The system prompt enforces an executive B2B framework (referencing specific initiatives, aligning value propositions, and closing with low-friction CTAs) while preventing hallucinated pleasantries.
Target Profiling & Sanitization (BeautifulSoup4 + Requests): The extract_website_context() utility issues clean GET requests disguised with browser headers, strips out navigational elements (<nav>, <footer>, <script>, <aside>), and condenses high-signal headings and paragraphs into clean contextual intelligence.
Studio Interface (Streamlit): The frontend features custom CSS, an indicator pill tracking local engine status, two-column campaign configuration, dynamic model selection, and tone toggles.
Data Export Pipeline (Pandas & urllib.parse): Allows instant lead export into .csv files and encodes multi-line email bodies directly into client-side mailto links.

Why Does Open Innovation Matter?

Open-source and open-weight AI matter because they level the playing field for independent builders.

If cutting-edge AI is restricted behind closed enterprise APIs, only funded startups can afford the token bills and subscription tiers needed to run autonomous sales workflows. Open models like Qwen 2.5 and LLaMA 3.2 prove that efficient, smaller open-weight models running on consumer hardware can match or outperform cloud APIs on deterministic tasks like JSON formatting and short-form B2B copywriting.

Furthermore, open innovation delivers:

True Cost Predictability: When you run a side hustle, your margins matter. Local inference costs ₹0 in API fees, whether you generate 5 emails or 500 emails a day.
Absolute Data Sovereignty: You aren't training proprietary closed-source models on your company's unique value props or your client's contact books.
Portability & Modularity: Because Ollama exposes an open, reproducible interface, any developer can fork this repo, swap models, or run it on their own hardware without an internet connection.

My First User

To test ColdPulse Studio in the real world, I used it to draft cold outreach for an EdTech consulting initiative aimed at Indian engineering institutions (such as NIT Manipur).

Using the app, I scraped the institute's website, generated an academic partnership pitch centered on curriculum-industry labs, and tested the resulting draft[cite: 1]. The feedback was immediate: the email avoided generic fluff, directly referenced recent departmental initiatives, and saved roughly 15 minutes of manual website research and drafting per contact.

BioScavenger

3078 Richa Kumari — Sun, 11 Oct 2026 15:52:26 +0000

BioScavenger: Bringing "AI Belongs to Everyone" to the Great Outdoors | Hacktoberfest 2026 Submission

Welcome to my official Hacktoberfest 2026 DEV Challenge showcase!

With this year’s core theme centered on "AI Belongs to Everyone"—moving beyond superficial documentation tweaks toward practical, open-weight, locally accessible artificial intelligence—I built BioScavenger[cite: 2].

As developers, we spend hours staring into IDEs, command lines, and glowing monitors. BioScavenger was built with one goal: leveraging offline, private, open-source AI to help programmers step away from screens, head outside, and touch grass[cite: 2].

1. What is BioScavenger?

BioScavenger is an open-source mindfulness companion and sensory field quest engine[cite: 2]. Instead of keeping your eyes locked onto a smartphone display, it generates location-aware outdoor objectives, synthesizes a private voice walkthrough, and prompts you to pocket your phone and explore nature using your five senses[cite: 2].

[ SELECT ENVIRONMENT & TIME ]
│
▼
[ OLLAMA: Qwen2.5 (3B Local LLM) ]
Generates 3 Sensory Nature Challenges
│
▼
[ PIPER TTS (Local Speech) ]
Synthesizes Calming Audio Guide
│
▼
[ POCKET DEVICE & EXPLORE IRL ]
│
▼
[ UPLOAD PHOTO PROOF & EARN ]
+50 Nature Credits per Objective

2. Core Architecture & Open-Source Stack

In direct alignment with Hacktoberfest 2026's open-weight AI mission, BioScavenger uses zero proprietary cloud APIs, eliminating costly paywalls and third-party data tracking[cite: 2]. The entire generation and speech pipeline runs 100% locally on your machine[cite: 1, 2].

Frontend & UX (Streamlit)
Powers the interactive interface, handles multi-column quest configurations, renders audio players, and dynamically tracks state via the Nature Exploration Score[cite: 2, 3].
Edge Intelligence (Ollama + qwen2.5:3b)
Executes local inference via Ollama's REST API endpoint with strict JSON schema constraints to dynamically compose 3 environment-specific challenges and a mindfulness script[cite: 2].
Local Neural Speech (Piper TTS + en_US-lessac-medium)
Ultra-fast, on-device ONNX neural speech synthesizer orchestrated via native subprocess streaming to compile real-time, zero-latency audio walkthroughs (quest.wav)[cite: 1, 2].
Runtime & Orchestration (Python 3)
Drives standard HTTP communications with the local LLM daemon, manages file I/O for photographic field evidence, and binds the stack with minimal external dependencies[cite: 2, 3].

🛠 Why This Architecture Matters for Open Source:

True Local Inference: By deploying Qwen 2.5 (3B) through Ollama, anyone with a modern consumer laptop can run sensory quests without a high-end GPU or cloud subscriptions[cite: 2].
Zero Latency & Complete Privacy: Piper TTS generates high-fidelity voice tracks in seconds using local ONNX weights—ensuring user prompts, location preferences, and audio walkthroughs never touch an external server[cite: 1, 2].
Modular Codebase: The backend cleanly separates quest generation, TTS rendering, and file documentation into independent steps, making it easy for contributors to drop in alternative local models or vision verification modules[cite: 2].

3. How It Works (Step-by-Step)

Step 1: Configure Your Expedition

Select your physical setting (Urban Park, Backyard / Garden, College Campus, Balcony / Rooftop, or Forest Trail) and choose your timeframe (10 to 30 minutes)[cite: 2].

Step 2: Local AI Generates the Quest

BioScavenger queries a local qwen2.5:3b model via Ollama with a strict JSON contract[cite: 2]. The model invents 3 tactile challenges tailored to the environment (e.g., inspecting specific bark textures, finding wild blossoms, or tracking wind currents)[cite: 2].

Step 3: Screen-Free Audio Briefing

The generated narrative is piped directly into Piper TTS, producing an immediate spoken field briefing (quest.wav)[cite: 1, 2]. You plug in your earphones, press play, lock your screen, pocket your phone, and head out[cite: 2].

Step 4: Field Verification & Credits

Once you return from the outdoors, document your findings by uploading photographic proof into the field journal to unlock +50 Nature Exploration Credits per completed objective[cite: 2].

4. How BioScavenger Embodies "AI Belongs to Everyone"

Hacktoberfest 2026 demands meaningful AI that is open, private, and accessible to anyone with standard hardware:

100% Free & Open-Weight: Runs completely offline without external API subscription costs[cite: 2]. Anyone with a personal computer can clone, run, and experiment with the system[cite: 2].
Private & Secure: No location prompts, audio syntheses, or personal journal photos leave your local machine[cite: 2].
AI for Human Well-Being: Demonstrates how generative models can reduce screen fatigue and encourage real-world mindfulness rather than driving endless digital consumption[cite: 2].

5. How to Run Locally

Get the application up and running on your local machine in under 5 minutes:

1. Clone the repository

git clone https://github.com/Richs03/BioScavenger.git
cd BioScavenger

2. Install dependencies

pip install -r requirements.txt

3. Pull the local open-weight model via Ollama

ollama run qwen2.5:3b

4. Launch the application

streamlit run app.py

Essential Links & Repository
GitHub Repository: https://github.com/Richs03/BioScavenger

Hacktoberfest Hub: https://hacktoberfest.com

Supported Models: Qwen 2.5 (3B) via Ollama & Piper TTS ONNX[cite: 1, 2]