Mar 2020 – Jul 2025 · 5 yrs 4 mos

• Managing a team of 7 members and experience of managing different variety of Projects for Bixby and Galaxy AI use cases.
• Developed NL modules for ASR Text Processing including Inverse Text Normaliser, Punctuation, NL-End Point Detection, etc for 8 lanuages including English, Hindi and Turkish.
• Developed Server based LLM pipeline to support 4 Galaxy AI use cases including Translation, Correction, Summarisation and Tone Conversion with Google Gemini.
• Developed On-device LLM deployment pipeline to support large models on S24 devices using Pruning, Quantisation, Factorisation and Mobile Runtime using MLC.
• Developed LLM Evaluation framework to evaluate LLM responses quantitatively & qualitatively across multiple criteria, e.g. Fluency, with minimal reference data.
• Expanded Bixby NLU pipeline to 2 additional languages for 4 devices while improving the In-scope(2%) and Out-of-scope(8%) accuracy reducing Data Drift.
• Incorporated Agile Methods to improve team productivity and quality by 25%
• Improved Coverage of NL classifier by developing BERT-based text classifiers while improving the latency, accuracy and data requirements for Training.

Mar 2019 – Feb 2020 · 11 mos

• Working on Human Robot Interaction AI Development, developed On-device solutions for Voice Interactions, Navigation and Action Planner based on Hybrid Rule-based and Deep Learning based Englines.
• Designed and Developed Multi-Modal NLU, Dialog Manager and Multi-Task Planner module for Samsung Robots in Education and Retail setup
• Developed OnDevice Multi-Modal Interpreter with late fusion method for Text, Face, Object and Touch modalities.
• Developed Navigation Module for Guiding & Following Customer use cases for Samsung Robot Challenge,2019 competing against Global centres.

Mar 2017 – Feb 2019 · 1 yr 11 mos

• Working on Bixby NL Platform Development, developed On-Device solution for Voice Assistant, researched and implemented Text Classifier based on Deep Neural Networks for 60+ Apps supported in Bixby.
• Improved Word Embeddings using character features to solve unknown word feature problem frequently observed in text classifier for User input based NLU systems.
• Lead a team of 3 Data Science Interns to design and develop a Multi-Modal Object Recognition module with Image and Test with research in various aspects of Early Fusion and Late Fusion.
• Designed and Developed the light weight NLU Engine and Machine Learning Models for Android devices
• Took responsibility of E2E project development of On-Device Bixby project including Scrum, SCM, etc.

Mar 2016 – Feb 2017 · 11 mos

• As a Senior Software Developer in the Samsung SVoice NLU Team, worked on commercialization of SVoice, while also working on pluggable side projects for improvements of existing NLU system.
• As a part of Hybrid NLU Task Force of 12 members from 3 ORC’s in Korea, Designed and Developed the Statistical-cum-Rule Hybrid NLU Engine; Took Ownership of Rule-based Engine & Integration, Design of NLU module
• Took charge for Integration, SCM, Deployment and Testing Management during Task Force duration
• Working in Intent Recognising Module of Rule-based Engine, Implemented parallel processor for matching of rules
• As a part of commercialisation work, took ownership of entry level domain classifier for rejection of utterance
• Achievements :
• Promoted to Lead Engineer in one year by fast promotion for excellent performance during Senior Software Engineer
• Received Employee of the Month for Sep’16 for Owning the requirements in Hybrid NLU Engine Development

Jul 2014 – Feb 2016 · 1 yr 7 mos

• As a Software Engineer in the Samsung SVoice NLU Team, worked on commercialization of SVoice, while working on side projects for improvements of existing NLU system.
• Developed a Multi-level text classification system using SVM’s to understand the user’s intent
• Using libSVM, experimented with rejection of out of context utterance using ensembles of classifiers
• Using Adaptive-SVM, designed Intent-Level Classifier to support Cross-Domain scenarios with faster training
• Mentored a team of 3 Interns to design and develop a State-based NLU engine with efficient Dialog Manager
• As a part of commercialisation work, maintained entry level domain classifier for rejection of utterance for Samsung S Voice Product.
• Achievements :
• Received Employee of the Month for Nov’14 for work done in development of Multi-level Text Classifier

May 2013 – Jun 2013 · 1 mo · Noida, India

• Android Development Team of Shifu, time management application for Android devices
• •• Worked extensively on Evernote API to add features for Call Notes, Reminders, Note suggestions, etc.
• •• Integrated LinkedIn with application and analyzed user data to create appropriate suggestions
• •• Developed product website and product log management interface to monitor product usage
• • Referred and Integrated RESTful web services (JAX-RS) to handle web requests from clients devices

Apr 2013 – Mar 2014 · 11 mos

• Spearheaded a team of 15 Hostel System Administrators and 30 Computer Secretaries
• • Monitored well-functioning of Institute’s network, IT infrastructure including Gymkhana servers
• • Ensured Speedy setup of network facility in newly established hostels in institute

Apr 2012 – Mar 2013 · 11 mos

• Nominated member of Hostel Affairs council to manage web related work involved in council
• Revitalized online Complaint Management System to cater complaints of over 8000 students
• Revamped Lost n Found and Guest Accommodation Booking System Website
• Developed Online Institute Market Place, buy/sell web interface and Student Affairs Website
• Introduced Election Website to conduct elections in 15 hostels each catering over 40 candidates

Dec 2012 – Dec 2012 · 0 mo · Mumbai Metropolitan Region

• • Improved Server Response time by implementing Database Caching & Object Caching (in File & RPC)
• •• Performed Front End Optimization tasks using Lazy Load/Loss-less Image Compression & JS Compression/Sub-domain Content Distribution methods
• • Designed and Implemented New module for generating Automated Purchase Orders in Back-end ERP for Operations
• Technologies:-
• LAMP (Linux Apache MySQL PHP), Code-Igniter Framework, Php Libraries (Caching & Image Compression)

May 2012 – Jun 2012 · 1 mo · Mumbai Metropolitan Region

• 24x7Style is an e-commerce venture for personal care and beauty products
• • Designed web-based interfaces with clean UI for support staff to ease process of order management
• • Included features of Data Ordering, Date Picker and Data Search with auto-complete