Day 100 of #100DaysOfCode
Wrapping my journey with the note that
✓Discomfort signals you're learning something real
✓ Pride in your code means you're ready to throw it away
✓ Laser sharp focus is necessary but not sufficent condition .
✓ ai tools are your collaborator not competitor .
thanks
day 99 of #100DaysOfCode
ideating for hackthon in civic tech space
day 98 of #100DaysOfCode
connecting the discord bot with the openclaw ...
neeed to work more on model provider
day 97 of #100DaysOfCode
takes lot times to install Openclaw in the device .
Started exploring it .
day 96 of #100DaysOfCode
using transformation of log, sqrt or boxcox to scale the values of feature to uniform range .
Day 95 of #100DaysOfCode
Built intuition for feature selection on breast cancer dataset 🎯
Calculated Fisher ratios for 30 features
Top separator: concavity_mean (3.49x)
Bottom: fractal_dimension_mean (0.0003)
Learning: Statistical separation > correlation heatmaps
day 94 of #100DaysOfCode
-Multicollinearity trap (radius~perimeter r=0.99)
-Outliers via IQR: kept them – extreme values = sick patients, not noise
-Class imbalance (62% benign) – RFQ preserves minority info best
-No data leakage;
Medical ML: outliers ≠ errors, they're clinical signal.
day 93 of #100DaysOfCode
Plating more with the medical dataset . Journey till now documented with the help of claude ai :
claude.ai/public/artif...
Day 92 of #100DaysOfCode
Medical ML isn't just accuracy. Learned my 90s breast cancer dataset has:
1) population bias
2) low events per variable (7)
3) high multicollinearity.
Small sample = unstable coefficients.
Context is everything.
day 91 of #100DaysOfCode
Getting started with breast cancer dataset .
Deep dive with the implication of false postive, false negatives and ways to mitigate them ..
Lowering the threshold and using at as filter and makeing biospy as fianl safety net is the only options .
day 90 0f #100DaysOfCode
I vow to not to miss anydays from now . If i miss single days , i am gonna start over counting from the begining .
Todays just preparing to attack the classification problem by taking hands on dataset of Breast Cancer Wisconsin.
day 89 of #100DaysOfCode
expanding the use case of my programming language . using python script to combine the episodes of web series into single one long file .
day 88 of #100DaysOfcode
expanding usecase of claude to genrate and refine the long resume content with compact one pager .
Helped a friend for this .
Day 86 and day 87 of #100DaysOfCode
Busy on crating and deploying my app polytinder .
you can try it at : polinder.netlify.app/swipe
Day 85 of #100DaysOfCode
Decided to go with citizen manifesto matchmaker for hackathon project , targeting marketing grade illustration .
besides this , revisiting data science concept and illustration that i implemented in the tips dataset .
Day 84 of #100DaysOfCode
Playing with Google gemini 3 and trying to ideate for Hackathon project .
Day 82 and 83 of #100DaysOfCode
Was such a super busy that I didn't update of yesterday also .
I created and deployed E-spprts analytics with data driven insights on cloud9-skylimit-hackathon-1.onrender.com
Source code :
Spent my day 81 of #100DaysOfCode on Assisted Pattern Recognition Coaching App with follopwing tasks :
✅ Refactored core modules using Clean Architecture & SOLID
✅ Improved UX: sticky header, better navigation flow
✅ Added data visualizations (charts + maps) for clearer insights
Forget to update yesterday.
Day 79 and 80 of #100DaysOfCode
Was stuck on the Junie not able to execute terminal command, later found that terminal was not configured for ide.
Just being comfortable with Junie ai in intelij IDE by telling it to refactor code to align with Solid principles.
day 78 of #100DaysOfCode
Generated the simple mvp to analyze player mistakes from match data to predict impact on team outcomes.
The match data is accesed through grid api .
please check the video
day 77 of #100DaysOfCode
no coding today . Just sumbitted my ernie baidu uid for collection of my prizes and certification ,
day 76 of #100DaysOfCode
Today , I cmpleted end-to-end baseline modeling on tipping data—from data cleaning to feature engrng, model selection, residual diagnostics, and coefficient interpret—achieving a 10.5% imprvment over naïve baselines with documenting thresholds for future comparisons.
Summary of my journey till day 75 of #100DaysOfCode
-Core ML concepts with PyTorch and completing key Kaggle courses
-Few projects built under civic tech themed
-Participated in hackathon like Kiroween and Hexathon
-Hands-on experience in data science through deep analysis of a real dataset.
day 74 of #100DaysOfCode
Just completed a ruthless model evaluation phase. I interrogated residual checks, segment analysis, Q-Q plots and Dollar impact and diagnosed failure modes.
At last , I translated the findings into waiter-friendly rules.
day 73 of #100DaysOfCode
just unlocked my ultimate ai pack of jetbrains to get participate in the sky limit hackathon .
Will start exploring its feature and tools from tomorrow .
day 72 of #100DaysODCode
sumbitted my project for hexathon .
🇳🇵 Analyzed 28 years of Nepal's migration data. Found a paradox: remittances grew 58× per capita, lifting incomes 12.6×—but creating structural dependency on foreign labor markets. We're being lifted up AND left behind.
#HexAThon #DataScience
app.hex.tech/virtual-hack...
day 71 of #100DaysOfCode
Used hex platform to analyze the migration platform of nepla .
have thought of making final submiison on the hackathon .
Day 70 of #100DaysOfCode
Getting comfortable with hex platform to do ai driven data analytics.
Learning to construct stories from data .
day 69 of #100DaysOfCode
Contuing working with a tip prediction model with 7 features, then discovered 5 had severe multicollinearity (VIF > 10) 😱
Pruned to 5 independent features. Result: clean coefficients + competitive accuracy.
