profile-pic

Ujwal Tewari

I’m Ujwal Padam Tewari, Senior Research Scientist and have an experience of 7+ years in Machine learning, AI and Generative AI technologies with core focus on building customer facing products.

I specialise in the real world deployment of AI-based models for automated control decision processes using both GenAI/RL/ML and evolutionary strategies in cases where human intervention would otherwise be difficult.

My work spans across:

• Building multilingual chat-bots with language detection/translation, intent classification and menu→static routing to reduce unnecessary LLM calls

• Shipping real-time LLM conversational AI evaluations with Hallucination, Relevancy and Completeness metrics for respone quality assessment

• Multi-agent RL for traffic intersections and HVAC optimisation with LSTM + RL

• Goal Oriented Dialog systems

• Localized Recommender units for Multi-modal inputs

With a proven track record in application oriented publications at NeurIPS Deep-RL workshop (2019) and ITSC 2020 and recent submissions to CODS-2025 and EMNLP 2026 ARR.

Highlights include “GameGuard”, a model-agnostic safety system with 99% jailbreak reduction, and

state-wise CPM/FDA models that delivered +5.94% LTP while removing third-party enrichment and

saving Rs 70 lakh/month

I have also been working as an AI Ambassador at Intel Corporation, a classroom mentor for the Deep Reinforcement

  • Role

    Senior Research Scientist-2 | Autodesk Revit developers

  • Years of Experience

    6.5 years

Skillsets

  • QLoRA
  • LLaMA-4
  • Llama-guard
  • LoRA
  • Markdown
  • Nova
  • Ollama
  • ONNX
  • OpenCV
  • PySpark
  • Python
  • Llama-3
  • Redis
  • Scikit-learn
  • Sonnet
  • Streamlit
  • Unsloth
  • XgBoost
  • Qwen-3
  • Grok-mini
  • Gemma-3
  • Co-pilot
  • Boto3
  • TensorFlow
  • AWS
  • Docker
  • FastAPI
  • GCP
  • Git
  • Linux
  • SQL
  • Transformers
  • Bitsandbytes
  • PyTorch
  • Cursor
  • Flask
  • GitLab
  • GPT
  • Haiku
  • HuggingFace
  • Java
  • Keras
  • LangChain
  • LangGraph

Professional Summary

6.5Years
  • Nov, 2022 - Present3 yr 3 months

    Senior Research Scientist-2

    Games24x7
  • Nov, 2021 - Nov, 20221 yr

    Senior AI Researcher

    Salesken.ai
  • Jul, 2019 - Oct, 20212 yr 3 months

    Research Professional

    Siemens
  • Apr, 2019 - Jul, 2019 3 months

    Reinforcement Learning (RL) Researcher

    Adventum Advanced Solutions

Work History

6.5Years

Senior Research Scientist-2

Games24x7
Nov, 2022 - Present3 yr 3 months
    Led ML and GenAI initiatives across Rummy & Fantasy - Agentic chatbot and its safety & evaluation; user journey personalization via campaign optimisations, delivering +5.94% user acquisitions growth coupled with AI driven personalized customer support. Architected & deployed an agentic microservices chatbot using Langraph with multilingual support. Built in-domain/out-of-domain guardrails (HingBERT-based) model coupled with LLM based user query rewrite before RAG. Designed a model-agnostic safety system (GameGuard) achieving 99% jailbreak reduction. Fine-tuned LLama3.2 7/8B, Qwen3 1.5/4/8B and Gemma3-4B using QLORA for profanity/jailbreak detection and dialogue response quality evaluation. Built state-wise XGBoost models for Conversion Probability Model (CPM) & First Deposit Amount (FDA) across Rummy and Fantasy. Implemented CPM and FDA segmentation to deliver state-specific personalized tile sets; trained Reinforcement Learning models on 1-year A/B experimental data to optimize tile amounts for each bucket/cohort. Impact: Growth of +5.94% user acquisitions, +4.3% revenue per user across platform, +15% growth in lifetime value of user; removed third-party, saving Rs 70 lakh/month.

Senior AI Researcher

Salesken.ai
Nov, 2021 - Nov, 20221 yr
    Senior researcher leading a team of Machine Learning engineers. R&D work on conversational-AI for sales context using Haystack reader-retriever system for query prediction in DB for cue generation in a real-time conversation for Customer service agent. Real-time smart classroom video pipeline for person attention and facial feature tracking. Multi-head attention with RL model development for hardware efficient Video summarization for video calls and classroom CCTV streams.

Research Professional

Siemens
Jul, 2019 - Oct, 20212 yr 3 months
    Key researcher in Applied Reinforcement learning for research at Siemens with work related to the development of real-time deployment of Object detection and RL models for traffic intersection analysis and congestion reduction. Pruning YoloV3 to decrease inference time on Nvidia Jetson. Energy and comfort optimization of buildings for Dubai Expo -2020 using LSTM based forecasting and RL based optimization for HVAC units. Autonomous drones and intelligent landing of the UAV patent filed.

Reinforcement Learning (RL) Researcher

Adventum Advanced Solutions
Apr, 2019 - Jul, 2019 3 months
    Key researcher in Applied Reinforcement learning for research at Siemens with work related to the development of Medical Image Segmentation using Unet and Deep RL to automate the OCT (Optical Coherence Tomography) tests for medical segmentation. RL used to detect anomalies in scans and segmented masks. Regression used to learn pixel mapping of different layers in OCT scans from outputs of Unet and RL module.

Major Projects

2Projects

Video Analytics for Zoom calls and Smart Classrooms

    Project VM2 involving participant name and window extraction from Zoom call, video summary creation, real-time analysis in classrooms.

Vehicular Traffic Optimisation Service, Project AKILA

    Processing live traffic streams using deep learning to optimize traffic signals, deploying RL models for intersection management.

Education

  • B.Tech, CSE & ML

    Indian Institute of Information Technology [IIIT-Vadodara] (2019)
  • High School and Inter (10th and 12th)

    City Montessori School (CMS) Lucknow