Archit Mangrulkar

235 Sanjana Park, near Agrawal Public School, Indore (MP), India 452016
 91- 8305414601 ! [email protected] è LinkedIn ¥ GitHub
Education
Indian Institute of Technology Kharagpur Nov 2020 – May 2024
Bachelor of Technology in Computer Science CGPA 9.18/10.00
New Greenfield Public Academy 2020
Higher Secondary 95.6%

Research Interests
Interplay of Computer Vision & Natural Language Processing, Bias & Explanability
Conferences & Competitions
The 29th International Conference on Computational Linguistics Oct 2022
7th workshop on Social Media Mining for Health Applications (SMM4H) | ACL | GitHub Gyeongju, Republic of Korea
∗ Co-authored and presented paper on Stance & Premise Detection shared task at the SMM4H workshop 2022
∗ Achieved a F1-score of 0.636 for stance classification subtask using the CovidTwitterBERT model, 0.664 for premise
classification subtask using BART-base model on the test set and the first position on the leaderboard
∗ Experimented with Parts of Speech (PoS), dependency parsing, Tf-Idf features and performed contrastive pre-
training on our best models using a supervised contrastive loss function

DevRev’s Domain-Specific QA Challenge Dec 2022 - Feb 2023

Silver Medal | Inter-IIT Tech Meet 11.0 | PS | Report | GitHub Kharagpur, India
∗ Developed Closed Domain QA system on SQuAD-like Datasets and generative augmentations using T5, GPT3
∗ Pioneered sentence-level improvisation to Facebook’s DrQA retriever to improve retriever latency by 3x
∗ Used FedAvg algorithm, Reptile meta-learning, Incremental Replay Mechanism to handle Domain Adaptation
∗ Achieved 0.85 F1 using Electra-BERT ensemble with 2.65x improved runtime using caching, ONNX, quantization

Sigmoid’s Emotion Detection Challenge Sep-Nov 2022

First Runner Up | Data Science Hackathon | PPT | GitHub Kharagpur, India
∗ Developed a multi-label emotion classifier for detecting 12 explicit & implicit emotions in tweets using SOTA models
∗ Implemented a strategy in which we encoded emojis, performed pre/post-processing, and used RoBERTa autoencoder
∗ Leveraged feature engineering by concatenating Parts of Speech (PoS) & dependency features with BERT encodings
∗ Achieved an F1-score of 0.59 with a union ensemble of three Roberta models implementing few-shot classification

DRDO’s (Ministry of Defence) UAV Guided UGV Navigation Mar 2022 - Apr 2022
Gold Medal | Inter-IIT Tech Meet 10.0 | PS | Report | GitHub Kharagpur, India
∗ Worked on Computer Vision aspect for developing an Unmanned Earth-Moving Ground Vehicle (UGV) and a guiding
Unmanned Aerial Vehicle (UAV) for mapping and navigating a mountainous terrain without collapsing
∗ Applied blurring, thresholding, contour detection to the UAV’s RGB-D image feed for UGV Localization
∗ Transformed UGV’s center’s coordinates in camera-frame using intrinsic camera matrix, UAV frame to local frame
and local frame to global frame transformation using Jijie Zhu’s algorithm to obtain GPS coordinates
∗ Employed Darknet-53 with YOLOv3 for estimating the UGV’s center but scrapped off due to non-generalized results

Research Experience
Computer Science Department, TU Dresden May 2022 – Jul 2022
Approximate Computing in Neural Networks | Prof. Akash Kumar Dresden, Germany
∗ Designed C++ based approximate multipliers using tensorflowlite for 8-bit inference of Multi-Layer Perceptrons
∗ Used 8-bit signed multiplier lookup tables, polynomial regression modelling in two variables and signed 8-bit multiplier
C-code files offered by the EvoApproxLib library to approximate the multiplication operator in different NN layers
∗ Improved inference times by 2x using OpenMP’s parallelizations, interchanged loop order and compiler optimizations
∗ Awarded the prestigious Erasmus+ Scholarship of over € 2900 for the duration of research traineeship
HCI-BCI Lab, IIT Kharagpur Nov 2021 – Present
Brain Connectivity Analysis | Prof. Debasis Samanta Kharagpur, India
∗ Worked on Temporal and Spectral aspects of Connectivity on the motor-imagery dataset using mne python library
∗ Computed statistical metrics such partial correlation coefficient, mutual information and canonical correlation
∗ Current work includes thresholding the adjacency matrices, performing graph analysis and implementing DL models
Chi-Square X Technologies Aug 2021 – Present
Execute ML-based client Projects | Machine Learning Engineer Kharagpur, India
∗ Authored blogs for European AI startup Connexun on Sentiment Analysis and Named Entity Recognition
∗ Worked on ML and DL classification models for preparing an Intrusion Detection System based on BoT-IoT dataset

Key Projects
Multi-class Document Classifier Sep 2022 - Nov 2022
GitHub | Tensorflow
∗ Developed a multi-class image classification model for classifying grayscale documents from the RVL-CDIP dataset
∗ Boosted accuracy score by 15% by appending a Supervised Contrastive loss function to the ResNET module
∗ Improved acc by 29% with bootstrap aggregation of MobileViT, MobileNet models achieving 77% overall accuracy

Instance Segmentation Module Jan 2022

GitHub | Pytorch, Tkinter, OpenCV, Matplotlib
∗ Created a package for transforming images and analysing their effects on the predictions of an instance segmentor
∗ Employed MaskRCNN model pretrained on COCO dataset that outputs segmentation masks, confidence & boxes
∗ Wrote plot-visualization utilities to plot segmentation masks and bounding boxes, implemented the GUI in Tkinter

Image Style Transfer Feb 2022

GitHub | PyTorch
∗ Implemented style transfer technique from the paper “Image Style Transfer Using Convolutional Neural Networks” to
generate using two images a new image that reflects the content of one but the artistic ”style” of the other
∗ Used SqueezeNet to extract features and used formulated loss function to perform gradient descent on the input

Deep Learning for IOT Oct 2021 - Nov 2021

Chi-Square X Technologies | Python, Keras, Tensorflow
∗ Developed a deep learning-based intrusion detection system (IDS) for the IoT based on the BOT IoT dataset
∗ Implemented Temporal Convolution Neural Network (TCNN) and Long Short Term Memory (LSTM) models
∗ Used Synthetic Minority Oversampling Technique-Nominal Continuous to handle unbalanced dataset
∗ Achieved accuracy score of 100.00% for attack prediction, 99.998% for category & 98.36% for subcategory prediction

Online Medical Consultation System Feb 2022 - Mar 2022

GitHub | Django, Python, React
∗ Developed a medical ERP website to manage medical consultation and appointments for patients, doctors & hospitals
∗ Implemented user features like booking and cancelling appointments, doctor features like accepting or rejecting ap-
pointments and admin features like removing doctors and hospitals
∗ Enabled Role based access control for stakeholders with permissions enabled on specific APIs at backend using Django

Position of Responsibility
Quant Club, IIT Kharagpur June 2021 – Present
Senior Quant Researcher Kharagpur, India
∗ Organized Summer of Quant, consisting of series of workshops and projects having a footfall of 300+ participants
∗ Co-authored blog on Quantitative Easing and presented Technical Analysis sessions covering candlesticks, indicators
∗ Worked on white paper presentations, creating alphas by developing and backtesting strategies using backtrader

Kharagpur Data Analytics Group, IIT Kharagpur Aug 2021 – May 2022
Associate Member Kharagpur, India
∗ Organised Kharagpur Data Science Hackathon which saw a participation of 2100 teams nationally
∗ Organised Python for ML workshop for students of IIT Kharagpur which had an attendance of of 250+ participants
∗ Co-authored blog on Decision Trees and delivered the Python & OOPs sessions in Python for ML Workshop
Relevant Coursework
Deep Learning: Stanford: CS229, CS224N, CS231N, DeepLearning.AI: DL Specialization, Jovian: DL with PyTorch
Mathematics: Probability & Statistics, Linear Algebra, Advanced Calculus, Numerical & Complex Analysis
Computer Science: Compilers, Systems Programming, Computer Organization & Architecture, Discrete Structures,
Formal Language & Automata Theory, Software Engineering, Algorithms- I & II, Programming & Data Structures

Technical Skills
Languages & Frameworks: Python, C, C++, gawk, LaTeX, Java, MATLAB, Django, Flask
Libraries: PyTorch, Tensorflow, Scikit-learn, NLTK, spaCy, HuggingFace, Matplotlib, Seaborn, Backtrader, OpenCV, PIL
Developer Tools: gcc, make, gdb, valgrind, gprof, grep, bash, Git, VS Code, Jupyter Notebook, MS Office

Academic Achievements
∗ Among the top 0.1% students to be offered Departmental change to Computer Science & Engineering at IIT
Kharagpur after securing 10.00 CGPA in theory subjects
∗ Secured an AIR of 978 out of 150k candidates in Joint Entrance Examination Advanced, 2020
∗ Secured an AIR of 1517 out of 929k candidates in Joint Entrance Examination Mains, 2020
∗ Qualified for Indian National Chemistry Olympiad in 2020 among the top 800 students out of 50k candidates
∗ Awarded Best Student in 2018 at Choithram School North Campus for excellence in academics