DEV Community: Tamanna Thakur

SocialSync AI: Channeling the Ghost in the Marketing Machine

Tamanna Thakur — Thu, 29 Oct 2026 05:38:11 +0000

Hi everyone! For the Week 4 Hacktoberfest Open-Source AI Challenge, I am back with an evolution of my original project, SocialSync! For this final sprint, I am interpreting the "Ghost in the Machine" theme by investigating the invisible, phantom parameters that alter our online voices.

I am building out the SocialSync Ghost Node upgrade—a spooky adaptation of our micro-agency dashboard designed to inspect and manage the underlying open-weight layers that automatically shape raw client text streams.

■ The Spooky Problem Stated
Every time a local micro-business uses automated generation tools, they are essentially letting a phantom algorithm take over their raw, human identity. If left unmonitored, this "ghost in the machine" strips out the organic soul of the message, replacing it with sterile, predictable patterns. Independent freelancers need a local workspace to inspect these phantoms before they haunt public feeds.
■ How the Ghost Node Illuminates the Pipeline
This upcoming sprint introduces an interactive tracking sandbox inspired by historic terminal interfaces, built around the framework of open-weight models (like Google's Gemma 2).
Phantom Token Isolation: The incoming intake screen lets a user paste a raw brief, utilizing client-side JavaScript logic to highlight the exact concepts the AI layer will alter, rewrite, or amplify.
■ The Offline Séance Architecture: Because it runs entirely as a static frontend layer hosted via GitHub Pages, the environment is designed to eventually run fully locally on a machine—meaning you can inspect these algorithm behaviors with the internet (and the lights) turned off.
■ Wrestling with these open-source layouts has been a highly creative milestone to wrap up the month, proving that we can study modern AI pipelines while keeping our deployment costs completely free and our data strictly private!

hf26challenge #opensource #webdev #ai

Sloth Brief — Open-Source AI for Lazy Automation

Tamanna Thakur — Sun, 25 Oct 2026 06:21:30 +0000

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

What I Built:

Sloth Brief — a deliberately lazy local agent that turns the messy stuff you keep putting off into a calm morning briefing. You drop group-chat exports, brain-dump notes, half-written emails, or any .txt/.md files into an inbox/ folder before bed (or whenever).

A small open-weight model running entirely on your laptop reads everything and produces a short Markdown file that contains:
Exactly two sentences that capture what actually matters
The three tiniest next actions (ideally < 2 minutes each)
A “safe to ignore until later” list:-
One low-energy win for the day:
The whole point is that the computer does the cognitive work so you don’t have to. You wake up (or finish your coffee) and already know the minimum viable next steps. No scrolling, no decision fatigue, no “I’ll deal with it later.”It is designed for a normal laptop, works fully offline once the model is downloaded, and never sends your private chats or notes anywhere.

Demo

Since Sloth Brief is designed as a local-first, privacy-focused agent (where private notes and chat logs never leave your device), there is no live web server required.

Source Code & Local Setup: Tam-ato/sloth-briefing

How to Run locally in 30 Seconds:

Pull the model via Ollama:


bash
ollama pull gemma4:31b-cloud   # or local gemma4:12b / llama3.2:3b 

Code
Key pieces:

src/sloth_brief.py – the agent (context gathering, prompt, local LLM call, optional watcher)

inbox/ – drop files here

out/ – generated briefs land here

examples/ – sample group-chat export + brain-dump notes

How I Built It
The core is a local open-weight model served by Ollama. I deliberately kept the architecture simple:

A few lines of code walk the inbox/ folder, newest files first, and concatenate contents up to ~1,100 characters.

A single system prompt tells the model exactly the Markdown shape it must produce.

The official Ollama Python library talks to the local server (localhost:11434).

Why Open Innovation Matters
Privacy by default: Personal notes stay strictly on your local machine.

Works completely offline: Once weights are downloaded, the tool works on a plane or without an internet connection.

Full control of behaviour: You can swap models, rewrite system prompts, or fine-tune without asking permission.

My Agent Session
Note: Developed locally using Ollama and Python CLI.

Building Process & Tradeoffs
Theme Selection (Slow and Steady): Built Sloth Brief to solve notification fatigue and info overload by turning scattered group chat exports and notes into a calm morning digest.

Local Model Focus: Chose gemma4:31b-cloud as default (with seamless offline fallback to gemma4:12b) via Ollama to keep sensitive daily notes completely private.

Context Truncation: Designed custom file-reading logic in Python to cap inputs at ~1,100 characters so local models run rapidly.

Structured Framing: Enforced strict Markdown formatting: 2-sentence context, 3 micro-actions (< 2 min), a safe-to-ignore list, and 1 quick win.

The Lazy Payoff
I used it myself while writing this post. Dropping two days of scattered notes produced a clean, actionable 30-second read.

Prize Categories
Best Use of Gemma: The project is designed around small open-weight models and defaults to gemma4:31b-cloud.

Week 2 Challenge

Tamanna Thakur — Sun, 18 Oct 2026 18:36:38 +0000

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

What I Built

It generates campaigns for brands trying to build their image and influence on social media.

Demo

https://tam-ato.github.io/socialsync-ai-dashboard/

Code

https://github.com/Tam-ato/socialsync-ai-dashboard

How I Built It

The application is architected with a decoupled frontend interface running entirely in the cloud using GitHub Pages. The core automation layout features:
• Intake Parsing Matrix: An interactive client-side JavaScript engine that wraps text token vectors.
• Open-Source Infrastructure: Designed to pipe local text streams natively through lightweight open-weight models like Google's Gemma-2B or Meta's Llama 3.
• Integrated Setup Environment: Configured and validated globally using the official DevRelay terminal tool to monitor session workflows and check system integration.

Why Does Open Innovation Matter?

Open innovation completely shifts the unit economics of a student-run side hustle:
• 100% Profit Margins: Traditional closed APIs (like OpenAI's GPT-4) impose strict per-request token costs that eat directly into a freelancer's margins. By shifting to free open-weight models deployed locally, operational overhead drops to zero.
• Absolute Data Privacy: Small neighborhood businesses are deeply protective of their promotional schedules, product launch menus, and internal notes. Running inference locally on an open-weight framework guarantees that sensitive commercial text files never leave the local laptop server or get stored by a third-party vendor.
• Zero Vendor Lock-in: Open-source architectures prevent sudden API price hikes or unexpected model deprecation, allowing students to swap model layers instantly as the developer ecosystem evolves.

My Agent Session

My First User

I shared the deployed static application links with a digital creator peer who handles content curation for local cafes. They pasted a chaotic text description of an upcoming weekend flash sale into the dashboard intake form. The application instantly converted the disorganized paragraph into a highly structured, emoji-mapped, copy-pasteable Instagram post layout. They noted that it completely replaced their tedious manual caption-drafting sessions and expressed a strong willingness to subscribe monthly once multi-account scheduling nodes go live!

Prize Categories

• Best Project Built Using Open-Weight Models (Gemma/Llama)
• Most Promising Open-Source AI Side Hustle / Micro-SaaS App
• Top Interactive Frontend Design & Deployed Static Wrapper

Building My Engineering Foundations: Hacktoberfest 2026 Learning Log

Tamanna Thakur — Fri, 16 Oct 2026 18:27:34 +0000

Hi everyone! As a second-year engineering student, I am using Hacktoberfest 2026 to dive into the world of open-source software and developer environments.

This week, my main practical achievement was configuring my developer workspace using the free DigitalOcean platform credits provided on the Hacktoberfest dashboard. In my university classes, I am currently mastering foundational data structures like stacks and queues, focusing on building clean, uniform display loops.

Connecting these code structures to modern open-source concepts is incredibly exciting. I have been researching how open-weight models (like Google's Gemma) are structured on the backend, and how developers deploy them using simple cloud servers. I am looking forward to applying these environment concepts to the upcoming live learning masterclasses later this month!

hf26challenge #opensource #learning #webdev