Divi Eswar Chowdary

AI/ML Engineer · Computer Vision · LLMs · Agentic Systems

Andhra Pradesh, India

LinkedInX
ED

About

AI/ML Engineer with expertise in computer vision, LLM-powered agentic systems, and production-grade deployment. Published researcher with hands-on experience building and scaling AI solutions across safety monitoring, enterprise intelligence, and NLP — from prototype to production.

Work Experience

Schlumberger
Onsite

July 2024 - Present

Data Scientist

  • Digital Factory – AI-Powered Safety & Operations: Engineered a computer vision platform monitoring compliance across 30+ facilities and 800+ cameras; fine-tuned YOLO models for PPE detection and forklift proximity alerts, deployed via TorchServe with torchao quantization for production inference.
  • Resolved false positives in lifting compliance by integrating monocular depth estimation to correct perspective distortions; replaced frame-level classification with VideoMAE video classification, improving alert accuracy from 70% to 90%.
  • People Finder – Enterprise Expert & Org Intelligence Agent: Co-designed an LLM-powered agentic system serving 1,000+ employees, enabling natural-language queries over reporting structures, skill directories, and org data; architected multi-step reasoning using LangChain and LangGraph with hybrid SQL + vector retrieval.

Schlumberger
Onsite

June 2023 - August 2023

Data Scientist Intern

  • Supply Chain Intelligence: Built a multi-task deep learning model generating unified product embeddings from RFM attributes and metadata, enabling alternative product recommendations for supply chain decision-making.
  • Designed a hybrid MLP-Transformer backbone applying self-attention over categorical and continuous RFM features; shared representations jointly optimized across multiple downstream tasks for robust per-product embeddings.

Education

Amrita School of Artificial Intelligence, Amrita Vishwa Vidyapeetham

2020 - 2024
B.Tech in Computer Science and Engineering (Artificial Intelligence) · CGPA: 8.77/10

Skills

Languages:
Python
SQL
ML/DL Frameworks:
PyTorch
TensorFlow
Scikit-learn
HuggingFace Transformers
timm
YOLO
VideoMAE
TorchServe
torchao
LLM & Agents:
LLM Fine-tuning
Model Merging
LangChain
LangGraph
RAG
Prompt Engineering
Tools & Deployment:
Docker
FastAPI
Streamlit
Gradio
HuggingFace Spaces
AWS SageMaker
GCP
Supabase
pgvector

Projects

LumaBot: AI Customer Support SaaS

Built a multitenant SaaS platform for deploying AI chatbots; architected a RAG pipeline with Qdrant, LlamaIndex, and OpenAI embeddings over a FastAPI + PostgreSQL backend with real-time streaming, async ingestion, and Docker deployment.

FastAPI
PostgreSQL
Qdrant
LlamaIndex
OpenAI
Docker
RAG

MedGPT: Medical Diagnostic Chatbot

Fine-tuned LLaMA-2 (7B) using QLoRA on a custom ~10K medical Q&A dataset built by scraping and translating Korean medical sources; augmented with a RAG pipeline for grounded, citation-backed responses.

Python
LLaMA-2
QLoRA
RAG
Fine-tuning

ModelHub

ModelHub is a platform for sharing, discovering, and running machine learning models.

Full Stack Developer
JavaScript
React
Node.js
Python

Podcastify

Convert articles from URLs into listenable audio podcasts.

Python
LLMs
Streamlit

RosBot: Self-Driven Robot with Obstacle Avoidance

Designed and implemented RosBot, a self-driving robot with obstacle avoidance, using ROS, TensorRT, Resnet18, and OpenCV. Optimized Resnet18 model with TensorRT for accurate obstacle detection and utilized OpenCV for efficient image processing.

ROS
TensorRT
Resnet18
OpenCV

Research Papers

An analysis of data leakage and generalizability in MRI based classification of Parkinson's Disease using explainable 2D Convolutional Neural Networks

2024
This paper presents a study on the detection of Parkinson's Disease (PD) from T1-weighted MRI scans using Convolutional Neural Networks (CNNs). The study investigates the potential for bias propagation in CNN models due to data leakage and evaluates the generalizability of the models to external datasets.

Transformer-Based Multilingual Automatic Speech Recognition (ASR) Model for Dravidian Languages

2024
This paper presents a approach to ASR for Dravidian languages like Tamil and Telugu. It builds upon the strengths of the powerful Whisper model, known for its multilingual capabilities, and fine-tunes it specifically for these under-resourced languages. This approach achieves significant improvements in WER compared to existing models.

Going Beyond Traditional Methods: Using LSTM Networks to Predict Rainfall in Kerala.

2024
This paper presents an approach to forecast rainfall in Kerala using LSTM models. The study compares the performance of LSTM models with traditional time series forecasting methods and demonstrates the superior accuracy of LSTM models in predicting rainfall in the region.