Passionate for Computer Vision and Data Analytics
Proactive Problem Solver | Curious & Excited for AI Era
Technical Skills: Python, Azure Vision,Intermediate-SQL, No-Sql,Google-cloud, Tableau, R, MLOPS, Deep Learning, Computer Vision, Kaggle
Analytical Skills: SOTA Ninja by following and using them in projects, Connecting with people to stay updated.
Education
-
M.Sc., Data Science(1st year) |
Technical University, Dortmund, Germany(October 2023) |
-
B.Sc., Applied Statistics and Analytics |
Devi Ahilya University, Indore, India (June 2022) |
Work Experience
Mostly worked on Freelancing projects, Also I did a lot of small projects for a lot of individuals, which couldn’t be mentioned here for privacy reasons.
- Data Scientist- MINDFUL AUTOMATION PVT. LTD.
- Arabic Billing OCR
-
- Extracting tables from Arabic Invoices, and using an OCR to scape detailsand output to excel tables and jsons. CHECK OUT PROJECT
-
- I used G-cloud-vision and G-cloud-translate api for the project. I also used Paddle-OCR(SOTA 2022) instead of pytesseract. It was a good learning experience.
- KYC
-
- Process of KYC(Know your customer) involves verification of official documents which consists of
sensitive information but involved human interaction. this sensitive information could get leaked, and so I
had to automate the whole process.
-
- I spent a good amount of time doing data pre-processing and augmentations, then I tweaked a lot of
algorithms to solve the problems. finally, I automated the process of verification. At first I used RCNN then Faster RCNN but hardly had any success with them.
-
- Finally, I used SOTA 2020 YOLO, to perform object detection, that gave me an accuracy of 88%, then I performed OCR which gave me an accuracy of 90+%.
- Invoices to Excel
- I was presented with the task of extracting tables from invoices using OCR.
- I Used Paddle OCR with detectron2 and GoogleVisionAPI to achieve the results.
- here is a link to the project.
- DATA SCIENCE CONSULTANT (SELF)– KERALA FREELANCER
- Sea weeds are medicinal herbs, but requires a lot of skilled man hours to harvest them. I was given the task to make a model to classify these weeds and harvest it using robots. But given the scarce availability of data, I was able to get 25 images from Client. CNN’s were overfitting and took 40 hours to converge with high BIAS.
- Finally made a DETR(data efficient Transformer) with a 60% accuracy.
- In the end we have to stop the project because of budget constraints in data collection, gained a lot of experience in that project.
-
(Intern)CURRICULUM DEVELOPER TECHNICAL CONTENT WRITER- MODERN DATA COMPANY MODERN DATA COMPANY
This is a product based start-up, My main responsibility was to create tutorials and write technical documentation for people wanting to use their product. I learnt a lot about Data Policies, wrote articles about data protection, polished my SQL skills, api-calls, wrote tutorials for developers.
- FREELANCE SELF
I wrote blogs, taught students Machine learning and Python accross the globe and consulted companies and individuals, with a team of 10 people who work with me and we grow together.
Talks
Projects
- CapiPort
Me and My friend made a tool for Portfolio Optimization of Indian Equity Markets. Our Main aim is to minimize Risk and Maximize Returns Of the portfolio, and properly allocating assets based on risk and returns.Link
- Kaggle Notebooks & Competitions-
I have wrote some good notebooks on Kaggle, I try to keep my notebooks as practical as I can. LINK.
- I teamed up with Samuel Courtinhas and came 32 🎉 out of 750+ in Kaggle Competition.
- Winner 🎉 of Data Grandmasters Event in my city. We competed against 20 teams, where we were asked questions related to analytics, statistics, and deep learning.
Deployed Gradio Projects @HuggingFace
my blogging website expired, shifting blogs from there may take time. Although traces of my blogs can be found on my LinkedIN and MEDIUM
Volunteering
Robin Hood Army Served 4000+ hungry people during lockdown in India, where people starved for healthy and clean food supply.