Baqir Khan Email & Phone Number

Remote

- Migrated old architecture, named classic, which had scaling limitation, to new, sophisticated micro-service architecture, named Lego, which could scale up to 10x than the previous one and used 1/(number of users per tenant) less data storage and API calls to salesforce.- Built watchers on Lego, which would trigger rebuild of a few tables, called materializers, when either a row in source table was created/updated or particular column in different materializer would change. Most challenging part about this was to avoid creating any infinite loops.- Onboarded all customers on classic to lego, a month before projected deadline, saving AWS cost for running old architecture, which was about ~$8,500.- Delivered many blueberries, tasks which engineers can pick while on call, that are good to have. One of them was adding search capability on one of our internal tools, which saved 70% of QAs team time while testing.- Led investigation of multiple SEV 1/2 incidents that directly impacted customers, used tools like datadog, rollbar to go back in time and find the root cause. Implemented quick fixes in record time along with scripts that cleaned up data, if needed.- Upgraded the writer service, responsible for writing data to salesforce, from using REST API to use Bulk API. This reduced the API consumption by (number of writes per Bulk API call, usually 200)/4 and reduced the load on the message queues & server by over 15%.

Remote

- Designed & built arch to calculate & store computed properties, asynchronously at the time of writes, with the help of previously built job system, Workman, hence eliminating need of joins and therefore fetching results in O(1) intead of O(n)- Enhanced Nudges further into boards, a web dashboard called Console. Boards are highly configurable dashboards, using globally standard lib such as Intl and Luxon, provide data asynchronsly in lazy loaded manner. Also built full fletched sidebar for console.- Dockerized entire application using docker compose, using Mongo, Postgres images, handling dependencies of services. Reduced developer onboarding time by over 90%.- Add-hoc tasks: - Improved error handling and logging using tools such as papertrail & datadog - Maintaining salesforce to application and vice versa sync, both bulk and incremental case - Contributing in open source project, Graffy, used by the firm, introducting things like query batching, increasing performance upto 70%- Took existing arch of generating prompts using setTimeout and made it distributed using multiple workers and message queue, stored state in NoSQL DB, hence improving overall reliability of the system, a.k.a Workman- Enhanced prompts further into Nudges, built APIs for serving data in real-time, calculated based on set of complex queries. Used indexing to improve performance by 60%.- Added functionality to send automated emails which included Nudge data, at a specific time,based on user's timezone

Gurgaon, Haryana, India

- Lead Vendor De-duplication and SKU Categorization projects, used by clients across US & South East Asia, improved accuracy with Edit Distance, Token match, Semantic match, etc. Exposed REST API's for backend-frontend On Demand integration.- Developed robust and scalable backend of Rene from scratch. Wrote complex queries in Elastic Search, reducing response time of every query to less than 200ms.- Added layer on top of Natty to developed robust date-parer to extract and process dates supporting 100+ unique ways of expressing dates.- Performed K-Means Clustering on Enron's Data-set of emails after expressing them to equivalent Google's Universal Sentence Encoder 512✕1 Dimension vectors using TensorFlow to meet objective of clustering emails which have similar theme or intent, after which, extracted relevant clusters based on fixed criteria and trained a classifier on it.- Using Deepmoji, an AI powered neural network trained on 1.5M tweets by MIT, converted chunks of texts into emojis after tokenizing them into paragraphs, and then, paragraphs to sentences using Natural Langugaue toolkit (NLTK) to filter out mails which convey specified type of ideas.- Resolved Co-references using Neural Coref to establish relationship between different sentences within paragraphs by replacing pronouns with intended proper nouns and nouns. Improved accuracy by 70% with the help of BERT (Bidirectional Encoder Representations from Transformers) + a complex layer of 1000 LOC on top of it.

Noida Area, India

-Coded Tree based solution for PPS & Anomaly identification models to compare & add rules to handle outliers for a fortune 500 firm.-Performed end to end deployment on production, handled web services and JQuery/UI related issues.-Automated indexing process on the database which reduced 2 hours/week of manual efforts.-Performed investigation to address client queries related to model performance and application specific change requests.

Bachelor Of Technology (B.Tech.), Computer Engineering

X: 10 Cgpa, Xii: 92%