This is a submission for the GitHub Finish-Up-A-Thon Challenge
we have all been there.
you stay up until two in the morning, fueled by coffee and a brilliant idea,
coding away during a weekend hackathon. you build a fantastic prototype,
but as soon as the event ends and your normal schedule takes over,
that project gets shoved into your github profile and forgotten.
it just sits there, collecting dust, waiting for the day you finally
have the time to make it complete.
i wanted to take one of my most ambitious, unfinished projects and
finally turn it into a polished, production-ready system.
a little about me
my name is ANIRUDDHA ADAK.
i am a final-year B.Tech student in Computer Science and Engineering
at the Budge Budge Institute of Technology (BBIT) in Kolkata, India.
over the last few years, i have worked as a freelance developer,
contributed heavily to open-source repositories,
and built autonomous ai systems.
i love combining modern frameworks with intelligent models,
but i also have a bad habit of starting massive projects
and leaving minor issues unresolved.
this challenge gave me the focus i needed to revisit skillsphere,
my comprehensive productivity and wellness ecosystem.
the mess inside skillsphere
i originally built skillsphere to break out of tutorial learning.
the idea was to build a single platform that could house ten different
mini-applications to help people improve their daily lives. the concept
was vast, hosting all of these tools together:
habit trackermood-based recipe recommendersustainable product comparisonpersonalized skill buildervirtual body language coachcrowdsourced travel recommendationsneighborhood micro-task exchangewellness companionar workspace plannerlive skill exchange network
while the design looked great on paper, managing ten different tools
inside a single workspace led to massive configuration errors.
the codebase was highly advanced, with typescript making up 97.8%
of the platform. however, if you looked at my repository files, you
would find a confusing contradiction.
the setup guide in my readme told users to run python package
installer commands, specifically instructing them to run
pip install requirements
this mismatch happened because i had tried to build local machine learning
tools directly into the frontend structure.
because of this, the environment constantly crashed during builds,
which made me put the project on hold. the main files like
tsconfig.node.json and package.json were filled with overlapping
dependency errors, and several apps like the ar workspace planner
remained half-finished.
rescuing my code with github copilot
i used github copilot as my guide to clean up the codebase and fix
my issues. my transition came down to three main steps:
step one: clean up the installation steps
copilot analyzed my codebase and helped me realize i could replace the
bulky local python scripts with clean client-side logic and
cloud-native ai calls.
this let me remove the confusing pip install instructions from the
setup guide, allowing new developers to deploy the workspace with
simple node commands.
step two: fix the configuration files
copilot helped me rewrite my tsconfig.node.json and package
dependencies to prevent vite build conflicts.
this fixed the structural errors that had been breaking my builds.
step three: polish the user interfaces
i used copilot to quickly generate the missing interactive layouts
for the ar workspace planner and the live skill exchange,
using tailwind css and framer motion to create
a unified experience.
my evolving project portfolio
completing this project represents a major milestone in my journey as
an ai engineer. to understand how my approach has changed, it helps
to compare skillsphere with some of the other applications i have
built over the years.
| project name | primary tech stack | original state | refactored state |
|---|---|---|---|
| skillsphere |
typescript, react, vite, tailwind css
|
stalled with incomplete sub-apps and a confusing python setup | fully functional suite of ten apps deployed seamlessly on vercel |
| lingolens |
next.js, assemblyai lemur, gemini api, tailwind css
|
basic speech-to-text prototype with simple translation features | comprehensive media analyzer with sentiment analysis and speaker tracking |
| homewhisper |
typescript, gemini ai, computer vision, vite
|
core voice commands and a simple hand gesture library | advanced predictive scheduling, safety diagnostics, and real-time tracking |
| voicemath |
react, typescript, tailwind css, assemblyai api
|
basic dynamic math quiz with simple voice response capturing | fully polished quiz application featuring a practice mode and leaderboard |
looking at this table, you can see a clear path of growth.
my earlier projects like voicemath relied on basic web voice
recognition apis to capture user responses and run simple quiz logic.
they were fun and highly interactive, but they did not have
deep processing capabilities.
with lingolens, i stepped up by using assemblyai's lemur api
and google's gemini to perform deep sentiment analysis and keyword
extraction on audio files.
finally, with homewhisper, i pushed the boundaries of multi-modal
systems by writing advanced hand-tracking computer vision algorithms
alongside context-aware voice commands.
this trajectory shows that i am shifting away from static web tools
and moving toward autonomous, multi-modal systems.
the value of building in public
participating in this challenge is not just about competing for prizes.
it is about building a habit of consistency.
during hacktoberfest 2024, i made over 238 pull requests to
open-source projects, which taught me the value of clean code
and documentation.
writing about my code on platforms like dev helps me understand
my own mistakes. it forces me to break down complicated patterns
so other students do not have to struggle through the same issues.
building in public keeps me accountable and pushes me to turn
my rough ideas into stable systems.
what lies ahead
finishing skillsphere has shown me how useful ai assistants are
for managing complex codebases.
as i finish up my computer science studies, my goal is to continue
diving deeper into agentic ai frameworks. i want to build systems
that can work together to solve complex problems, such as my
autonomous agent marketplace agentforge or multi-agent devsecops
systems like secureops-ai.
reviving this repository was a challenging experience, but it was
highly rewarding.
i have turned a confusing, half-finished repository into a clean
workspace that is ready for other developers to use.
Top comments (0)