Ayush Kumar

Lead ML Engineer

Bengaluru, Karnataka, India11 yrs 7 mos experience
Most Likely To SwitchHighly Stable

Key Highlights

  • Led ML team at Observe.AI for product maturation.
  • Published 20+ peer-reviewed papers in NLP and speech.
  • Developed innovative AI solutions for contact centers.
Stackforce AI infers this person is a SaaS AI/ML expert with a focus on conversational systems.

Contact

Skills

Core Skills

Machine LearningNatural Language ProcessingGenerative AiLarge Language Models (llm)Sentiment Analysis

Other Skills

AlgorithmsAmazon Web Services (AWS)Artificial Intelligence (AI)BloggingComputer ScienceDeep LearningEarly-Stage StartupsEmotion AnalysisExtract, Transform, Load (ETL)Feature ExtractionFine TuningGitJavaModel DevelopmentModel Training

About

**** Curiosity x Product x AI/ML Research x Strategy x Impact **** I have built AI systems for 0->1, 1->10 journey — with 10 years of industry experience in converting said and unsaid business problems into ML formulations and deploying AI/ML systems at scale for real-life human language (majority of experience in dialog systems and conversational domain), for products used by thousands of human agents and supervisors. My journey has woven through the evolution of machine learning: starting with classical models (SVM, LR, RF) and advancing through LSTM → BERT/Transformers → Large Language Models → GenAI Agents. Across this arc, my focus has always been on understanding how to represent/encode the data, where models fail in practice, why they fail, and how to pre-emptively address those failure modes so that systems deliver reliable impact at scale. As a Senior Manager at Observe.AI, a conversation intelligence platform that has raised $213M+ in funding and serves 350+ enterprises globally across industries, I’ve led the team of ML Engineers and Scientists along with data annotation group, who together contributed to the growth of ML applied research and product integrations from early team stages through organizational scaling and product maturation. My leadership has consistently synthesized strategic foresight, empirical rigor, and cross-functional execution — anticipating bottlenecks before they become risks and translating research into frameworks, tools, and insights that inform product roadmaps and engineering prioritization. Over the years, I’ve worn many hats across research and engineering, discovery, and strategic guidance— primarily influencing the product directions by bridging the art-of-possibility in AI with genuinely useful user experiences via AI products. Along the way, I’ve published 20+ peer-reviewed papers across premier NLP and speech venues (EMNLP, NAACL, Interspeech, ICASSP — including main, workshop, industry, and demo tracks), and contributed to 1 granted patent and 1 filed patent — all rooted in real conversational data and brought in innovation and differentiation that matter to users and enterprises.

Experience

11 yrs 7 mos
Total Experience
3 yrs 10 mos
Average Tenure
7 yrs 4 mos
Current Experience

Observe.ai

4 roles

Senior Manager, Machine Learning

Promoted

Oct 2024Present · 1 yr 6 mos

Generative AIPerformance AppraisalModel TrainingModel DevelopmentFine TuningComputer Science+26

Manager, Machine Learning

Apr 2022Sep 2024 · 2 yrs 5 mos

  • Building LLMs and GenAI solutioning for contact center use-cases
  • The Paradox of Preference: A Study on LLM Alignment Algorithms and Data Acquisition Methods (NAACL 2024 - Insights into Negative Results)
  • Can probing classifiers reveal the learning by contact center large language models?: No, it doesn’t! (NAACL 2024 - Insights into Negative Results)
  • Tweak to Trust: Assessing the Reliability of Summarization Metrics in Contact Centers via Perturbed Summaries (NAACL 2024 - TrustworthyNLP)
  • Investigating the Role and Impact of Disfluency on Summarization (EMNLP 2023)
Technical LeadershipGenerative AILarge Language Models (LLM)Startup EnvironmentsEarly-Stage StartupsAmazon Web Services (AWS)

Sr. ML Scientist

Promoted

Oct 2020Mar 2022 · 1 yr 5 mos

  • Building language models capturing natural conversational properties while being robust to ASR errors
  • PhonemeBERT: Joint Language Modelling of Phoneme Sequence and ASR
  • Transcript (Interspeech 2021)
  • Wearing multiple hats on bringing machine learning solutions for call center quality management process
  • Developing keyphrase suggestion model for discovering call events
Technical LeadershipGenerative AILarge Language Models (LLM)Startup EnvironmentsEarly-Stage StartupsAmazon Web Services (AWS)

Machine Learning Scientist

Sep 2018Sep 2020 · 2 yrs

  • Emotion and sentiment analysis for customer voice calls
  • Gated Mechanism for Attention Based Multimodal Sentiment Analysis (ICASSP 2020)
  • Customer intent detection
Startup EnvironmentsEarly-Stage StartupsAmazon Web Services (AWS)

Ipsoft inc.

Research and Development Engineer (ML & NLP)

May 2016Aug 2018 · 2 yrs 3 mos · Bengaluru Area, India

  •  Working in Episodic Memory team to add the conversational ability to the flagship AI agent – Amelia.
  •  Developed competing systems for retrieval based dialog system that cater to contextual understanding of the conversation and completion of goal.
  •  Worked on intent classification and textual similarity. Typically worked on deep learning systems involving CNNs, LSTMs, attention mechanisms, Seq2Seq encoder-decoder models, word and sentence representations.

Technische universität darmstadt

Research Intern, Language Technology Group

May 2015Jul 2015 · 2 mos · Darmstadt Area, Germany

  • I am associated with the project on 'Aspect Based Sentiment Analysis' under the supervision of Prof. Dr. Chris Biemann, Head of Language Technology Lab, TU Darmstadt.
  •  Designed a supervised system for sentiment polarity classification and aspect category identification based on domain dependency graph, distributional semantics and lexical features.
  •  Created sentiment lexicons for Hindi and Bengali language using the concept of distributional thesaurus and lexical acquisition.
  •  Bagged 1st rank in ‘Sentiment Analysis in Indian Languages’ task for Bengali and 3rd for Hindi using the developed system.

Indian institute of technology, patna

Undergraduate Researcher | AI-NLP-ML Lab - IIT Patna

May 2014May 2016 · 2 yrs

  •  Researched on the bottlenecks of sentiment analysis at different level of granularities using methods of: deep learning, active learning, multi-objective feature selection and lexical acquisition based on distributional thesaurus for aspect based sentiment analysis, contextual polarity disambiguation and message polarity classification.
  •  Proposed a hybrid deep learning architecture for sentiment analysis that beats previous state of art systems in resource poor languages.
  •  Our system is placed 1st in sentiment polarity classification for English laptop domain, Spanish and Turkish restaurant reviews and aspect extraction for Dutch and French in restaurant domain for the task Aspect Based Sentiment Analysis of user reviews.

Education

Georgia Institute of Technology

Master of Science - MS — Computer Science

Indian Institute of Technology, Patna

Bachelor of Technology (B.Tech) — Computer Science

Jawahar Vidya Mandir, Shyamali, Ranchi

+2 CBSE — Science Stream

D.A.V. Public School, Lalpania, Bokaro

Matriculation

Stackforce found 100+ more professionals with Machine Learning & Natural Language Processing

Explore similar profiles based on matching skills and experience