Work Experience #

• Working at Fora, a part of VerticalScope’s cloud-based digital platform that operates over 1,200 online communities connecting 110 million active users monthly, fostering their passions, and facilitating knowledge sharing.
• My role focuses on building healthier and better communities by empowering users with machine learning. Here are some of the tasks I have been responsible for:
  • Empowering communities with machine learning to enhance user experience.
  • Personalized Feed Recommendation’s, Newsletter growth, mobile notifications to all communities.
  • Developing advanced search algorithms for accurate, tailored results.
  • Optimizing product retrieval and facilitating new product discovery within communities.
  • Answering user questions and providing technical support, using LLMs.
  • Interviewing and mentoring junior engineers.

• Enhanced the in-house News Recommendation System using Sentence Embeddings.
• Identified and implemented Document Question Answering, as configurable classifiers to analysis various risks on businesses.

• Constructed and deployed an Entity Resolution System utilizing Unsupervised Contrastive Learning.
  • improved top-20 search accuracy from 35% to 98% for 10^7+ entities.
  • Cut down search latency from 1.5s to 0.3s.
  • Crafted a scalable system employing Postgres, Onnx, Pinecone, fastAPI, with Dockerized deployment.
• Conducted 35+ candidate interviews and developed a system for pinpointing the most suitable candidates for respective roles.

• Led the product development of Voicebot engine which powers voice-based queries on the MyAirtel app with 10m MAU, in 7 indian languages, does 500k queries/day. (Speech to text, Text to speech, training and inference pipelines.)
  • Voicebot integeration with PBX exchange like Asterix.
  • Presented work at Nvidia GTC Winter 2020
• Researched and deployed e2e OCR pipeline serving 1.6m docs/day at 96%+ accuracy, used by Airtel for its new customer acquisition journey ICDAR Rank 6
• Building the workflow-orchestration engine which powers the customer support queries on mail / social media for Airtel, processes 50k emails/day, built on k8, temporal.io
  • Supported 150 different workflows with ~50 activities running concurrently.
• Hired and led a team of 9 engineers.

• Built OCR pipeline for reading arbitrary documents, 5 step process with word localization, word recognition, clustering, parsing and serving.
• Built Voice based IVR bot for Indian business by building on top of DeepSpeech and Rasa NLU.
• Invited for YC 2018 Winter Interview Stage in San Francisco.

• Recommendation & ranking for users by clustering restaurants into latent topics space and recommending fresh restaurants built on node.js.
  • <200ms latency over a set of 1 million restaurants per user.
• Loyalty and rewards platform with second price ad bidding for banks (pilot run with Royal Bank of Scotland).

• Built the auto correcting & predictive completion language keyboards for regions like germany, japan etc based on the hidden markov model.
• Worked on Developing & Deploying Amazon Instant Video on 13 different living room TV environments in 10 months to 1m+ customers.
• Scaling & building a/b testing framework to test the application across various regions.

• Developed a JavaScript single templating based backend/frontend framework.
• Integrated native platform level code with existing Lua code using Foreign Function Interface.