Worked as trainee .
Worked on QlikView, built dashboards for marketing analytics.
Blockchain
Novel Consensus For Blockchain :
Objective : Design novel consensus for layer 1 blockchain for Better security. Better speed. Better nality.
Working on the implementation of a highly scalable and fast consensus engine in collaboration with some of
the world's top PhDs in Cryptography. Worked on Nested DKG, Tribe Clan Model for Consensus,
Bls Signatures. Worked on building L1 from scratch.
Technology : Rust
NFT Market :
Objective : Design an application that covers the creation of an interactive NFT marketplace where users can
buy, sell or bit NFT Collectibles.
Also, it has
Next Js Integration with Web3 and Blockchain.
Smart contracts are written in Solidity
Link : NFT Market
Marketplace application :
Objective : Design an application that covers the creation of an interactive marketplace. Clients of this
application will purchase courses students can choose products of their choice with Ether.
Page with real-time updates of Ether and course prices compared to the USD.
Also, it has
Next Js Integration with Web3 and Blockchain.
Creation of a more complicated Solidity Smart Contract.
Email hashing and order veri cation feature.
Testing solution for testing smart contract.
Pagination and Filtering of orders
Link : Courses MarketPlace
Decentralized Exchange for ERC20 tokens Dharma/ETH :
Objective : Create a decentralized cryptocurrency exchange DEx powered by Ethereum smart contracts
Designed frontend in React-Js for exchange.
Wrote smart contracts for depositing/Withdrawing and placing orders for Dharma Tokens.
Added Tests for testing features of smart contracts
Link : Kartavya Decentralized Exchange
ML Projects
Document OCR Python
Objective : to fetch information from documents by OCR
Done OCR on documents like PAN, Aadhar Front, Aadhar Back to give appropriate user-level data, including
photo
Used progressive calibration networks for rotation invariance to fetch photos from the document.
Developed an API for generating QR-code for UPI payments
Developed an API that will measure similarity between two names.
Developed an API that validates Aadhar no using the Verhoe algorithm.
Document Detection using Deep NeuralNets
Objective : Create a model that can identify the type of document from images
Developed a CNN model for identifying document types like PAN, AADHAR FRONT, AADHAR BACK, and
CHEQUE.
Use transfer learning ResNet50, added layers and trained it.
Was responsible for model training, creation and deployment.
Also created an API based on the Connexion framework.
Technology : Python, Connexion, Docker
Achievement : Saved huge cost for the company in doing preventing false OCR, saved operations cost
Cardiovascular Disease Detection Hackathon
Objective : to design a system for detecting cardiovascular disease.
Role - End to End Deployment, Testing, Monitoring, Retraining
Responsible for Detailed EDA, Model Creation, deployment, testing of model, monitoring and retraining
approach
Achievement : Obtained Recall of 82%, using F2 score as a metric.
Link :
https://github.com/vinay10949/CVD
SmartBackGround Veri cation using DeepNeuralNets
Objective : to use automate the problem of background veri cation.
Identifying eye descriptors and then calculating EAR I.e eye aspect ratio for the consecutive frame to identify if
the person in the frame is blinking his eyes or not.
Built Face Veri cation module to verify pro le photos with document photos.
Built Name & Address similarity engine which uses cosine similarity and Levenshtein Distance.
Short Term Loan Default Prediction Python
Objective : to create a model that can identify early loan defaulters.
Performed data engineering, data pipelining and detailed analysis of our customers.
Also, trained, tested model for predicting which rst time customer is likely to default 5000 Rupee product loan.
Achievement : achieved 0.79 F-Beta Score, loan default rate dropped by 40%.
Music Recommendation Engine
Objective : Solve the Cold start problem, increase engagement, Collect data of users on the music application
Used Music Information Retrieval MIR from the library by Vienna university Used features like rhythm pattern,
temporal descriptor, etc Approx 2000dimensions of one song Used Locality Sensitive Hashing algorithm
, Episilon Greedy Approach for recommendation
Technology : Python, Protobuf, Konga, Redis
Result : Solved Cold start Real-time music recommendation engine for use base of 50million users
BackEnd Developer Projects
Kartavya Delayed Queue
Objective : Increase e ciency of operations by delayed/real-time noti cations Close to 100Million
Noti cations per day
Delayed Queue Code Named Kartavya, is a smart queue which enqueue delayed actions inside the queue, then
performs action at scheduled Time.
Created Delay push message to the target, At-lease-once delivery, Fail and retry mechanism Queue is accessed
via 3 Rest API calls, to push message, to query schedule, delete a schedule, Added to take a snapshot to Redis.
Technology : Redis, Golang, AWS Cloud Result : Created the system with less cost and increased e ciency
Kyc Microservice
Developed a Microservice that encapsulates business logic for Video KYC of Karza and Signzy
Technologies : Golang
Data Pipelines
Built large data pipelines that migrate data from AWS Athena to BigQuery for analytics.
Used Python Celery for multiprocessing
API Gateway setup Kong
Was responsible for setting up API Gateway called Kong, Con gured all services and routes into Kong.
Con gured plugins for ACL, OAuth2, Request Termination.
Document Veri cation Service
Created Golang GRPC powered micro-service for extracting data from documents like Aadhar Pan, and voterid
using OCR.
Also created aadhar e-signing document ow for signing documents via NSDL.
Technologies : Golang