Sulabh Katiyar

CEO

Bengaluru, Karnataka, India4 yrs 2 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in Conversational Intelligence and NLP.
Developed innovative image captioning techniques.
Recognized for cost-effective model optimization.

Stackforce AI infers this person is a specialist in AI and NLP with a focus on multimodal systems.

Contact

Skills

Core Skills

Machine LearningNatural Language Processing (nlp)

Other Skills

Artificial Intelligence (AI)Automatic Image CaptioningBERT (Language Model)Computer VisionConvolutional Neural Networks (CNN)Data AnalysisData ScienceDatabasesDeep LearningDeep Neural Networks (DNN)DockerFastAPIInstruction Fine TuningKerasKnowledge Distillation

About

At Salesken.ai my work lies in the field of Conversational Intelligence. Here I am involved in research on semantic and cross-lingual textual relevance, knowledge distillation, language modelling and machine translation. I am currently pursuing PhD in Automatic Image Caption generation using Deep Learning Techniques and have recently submitted the thesis. My work is focussed on generation of single sentence descriptive captions for each image which capture the salient information in the image. My proposed methods for Caption Generation involve better extraction of visual information from images and multimodal fusion of visual and textual information. In addition I have also pursued the task of sub-region caption generation where a caption is to be generated for a pre-defined rectangular cross-section of the image using Hindi Language. Before pursuing my PhD I had worked on Abstractive Text Summarization towards my M. Tech thesis.

Experience

4 yrs 2 mos

Total Experience

1 yr 5 mos

Average Tenure

1 yr 4 mos

Current Experience

Eka.care

Lead Data Scientist

Feb 2025 – Present · 1 yr 4 mos · Bengaluru, Karnataka, India · On-site

Working on Medical Multimodal LLMs

Krutrim

Research Engineer 3

Jan 2024 – Jan 2025 · 1 yr · Bengaluru, Karnataka, India · On-site

Training, Alignment and Agents: Worked on both in-house LLMs (trained from scratch) and open-source LLMs (ranging in size from 100 million parameters to 400 billion parameters)
1. Research on Multi-Modal Large Language Models
2. Worked on Continued Pre-training of LLM.
3. Resource efficient Instruction Fine-Tuning of LLMs.
4. RAG: Training LLMs and Retrieval Models (Embedding models for ranking and re-ranking)
4. Multi-Agent Systems for Enterprise use cases. Building horizontal (covering wide range of domains) self-serve Agent Builder system.

Salesken

Research Engineer 1

Jan 2022 – Nov 2023 · 1 yr 10 mos · Bengaluru, Karnataka, India · On-site

Resource efficient language models
1. Cost-effective Knowledge Distillation to reduce model size by 40% while retaining 98% of model performance. Created distillation pipeline for all embedding models used in production.
2. Mini-sized Translation Models for Indian to English language translation. Model contains 90% less parameters than SOTA models for many-to-many translation. Received Quarterly Performance Award for this project.
3. Trained Generative Models for multiple use cases: Summarization, Objection Detection, Agent response, Intent discovery, Cues Generation, Call state classification and Pitch Intelligence. Created pruned versions of each model for efficient deployment.
4. Trained models for task specific use cases: Lead scoring, Semantic-similarity, semantic search, paraphrase generation, Intent clustering.
5. Finetuned Large language models (LLMs) like MPT-7b, RedPyjama-3b using PEFT techniques.

Knowledge DistillationLanguage ModellingMachine TranslationAutomatic Image CaptioningDeep LearningNatural Language Processing (NLP)+1