Pathikreet Chowdhury — ML Engineer & AI Researcher

Get to Know Me

About Me

ML Researcher & Engineer

As a postgraduate CSE engineer, I build production-grade AI systems that bridge research and real-world impact. I love to build code, research, and understand systems across the entire stack—doing everything from Frontend (FE) and Backend (BE) to Infrastructure and ML. My work spans LLMs, federated learning, computer vision, and large-scale data engineering, with deployments at companies like Flipkart and Siemens. I'm also the Founder & CEO of Artemis Pvt Ltd, building GenAI-powered safety solutions.

This portfolio is a reflection of my life—both professional and personal. When I'm not coding, you'll find me on the football field playing as a winger or centre forward, cheering for FC Barcelona, or competing as an international chess player. I'm also a proud cat parent to my cat, Jim Beam!

3 Publications

4 Companies

5+ Major Projects

Connect on LinkedIn

Where I've Worked

Work Experience

Data Engineer

Flipkart (Cleartrip) Bangalore

Industry

Architected scalable ETL pipelines with Apache Spark processing 200+ TB of travel data; improved pipeline execution efficiency by 30%
Advanced BigQuery optimizations — partitioning & clustering — achieved 25% cost reduction and 2× faster query performance
Reduced end-to-end data ingestion latency by 40%, enabling near real-time decision-making for Cleartrip's booking engine

Apache Spark BigQuery GCP Azkaban ETL

Research Engineer

Siemens AI Bangalore · Foundational Technology Research

Research

Integrated Physics-Informed Neural Networks (PINNs) with Symbolic Regression; reduced outliers by 52% on industrial datasets
Optimized Multi-Agent Symbolic Regression with LLMs in the loop; achieved 12% inference time reduction with improved accuracy
Developed Neuro-Symbolic Agents combining NSR with GenAI for molecular optimization and enhanced hypothesis generation

PINNs LLMs Symbolic Regression GenAI

Machine Learning Engineer

Matrice.ai Inc.

Industry

Integrated foundation models (SAM, GroundingDINO, CLIP, YOLO World) for video analysis with 30% accuracy improvement
Key contributor to the official Matrice Python package on PyPI; designed core architecture and reduced onboarding time by 45%
Developed custom CV solutions including gas station monitoring with multi-object tracking and hotel room similarity assessment

Computer Vision SAM YOLO World PyPI

SURGE Researcher

IIT Kanpur Under Dr. Vipul Arora

Research

Implemented Controllable Neural Symbolic Regression achieving 78% accuracy on a dataset of 230M equations
Adaptive Refiner-Based few-shot font generation with Monotype Inc.; reduced generation time by 40% while maintaining quality
Solved mode collapse in generative models with Normalizing Flows (RealNVP) and FAB; achieved 72.5% replication score

Neural Symbolic Regression Normalizing Flows Generative Models

Academic Contributions & IP

Research, Patents & Publications

Patent · Issued Nov 8, 2025

Automated Physically Informed Neural Network Enhanced by Symbolic Regression and Large Language Models with balanced Loss Optimization

Patent No: 2025E09468 IN

Invented a novel neuro-symbolic architecture integrating PINNs, Symbolic Regression, and LLMs for enhanced physical system modeling.
Engineered a balanced Loss Optimization strategy to efficiently unify physical constraints and data-driven accuracy.

ACCM 2026 · Under Review

LAST: Lightweight Adaptive-Shift Transformer for Real-Time HAR

Advisor: Dr. Gargi Srivastava

"More with Less" HAR architecture — shifts from dense video to sparse skeletal representations, achieving SOTA performance on edge devices
Developed Adaptive Graph Convolutions (A-GCN) + Temporal Shift Modules (TSM) with O(N) linear attention and Knowledge Distillation via VideoMAE V2

ICPR 2024 · Published

Brain Tumor Classification & Segmentation using Custom CNN and U-Net with XAI

Advisor: Dr. Gargi Srivastava

Custom CNN and U-Net for MRI diagnostics achieving 99.6% accuracy; introduced novel metrics quantifying LIME-based explanations for neuro-oncology
Fused Grad-CAM visualizations with interpretability scores, reducing false-positive rates in automated tumor detection

ICIT 2025 · Published

Federated Learning Framework for Privacy-Preserving Face & Activity Recognition

Advisor: Dr. Gargi Srivastava

FL architecture with adaptive aggregation; 65% bandwidth reduction via model pruning & quantization for decentralized surveillance networks
Differential privacy + adversarial feature disentanglement ensuring robust resistance against membership inference attacks

Research Project

An Efficient Machine Intelligence Framework for Detection of Heart Arrhythmia Disease using Cleveland Heart Dataset

Advisor: Late Dr. Rahul Kumar

Evaluated and integrated an ensemble of ML models after applying PCA and RFE, achieving 98.6% accuracy on test data.

BTP 2023

GestureCraft: A Novel Approach to ASL Sign Language Detection using CNN, LSTM and MediaPipe

Advisor: Dr. Nirbhay Kumar

Designed an efficient ASL detection system utilizing Innovative Dataset Capture Techniques, achieving 98% accuracy.

BTP 2024

Efficient In-Painting Method Using an Ensemble of Generative Models - DDPMs and GANs using Kaggle Little Images Dataset

Advisor: Dr. Nirbhay Kumar

Enhanced image restoration quality by 22% and minimized in-painting time by 30% using an ensemble of DDPMs and GANs.

Ongoing

Integrated Agentic Architecture with XAI for 3D Medical Scans Detection and Segmentation using BRaTS and LiTS Datasets

Advisor: Dr. Gargi Srivastava

Improved 3D medical scan segmentation and detection speed and real-time inference by 25% with UNET++ and LLMs, enhancing model transparency with XAI.

What I've Built

Featured Projects

Eir — Federated Learning Healthcare Platform

View Project

Eir

End-to-end Federated Learning healthcare platform with advanced RAG pipelines; LoRA-optimized SLMs achieving 45% faster training with HIPAA compliance and 3.2× faster inference on edge devices.

TensorFlow PyTorch LangChain LoRA CrewAI

View Project

LegalVoice.ai

Multilingual voice-to-text legal assistant supporting 8 regional languages at 92% accuracy; QLoRA fine-tuning and AutoGen agentic workflows reduced document processing time by 73%.

Whisper AutoGen QLoRA React RAG

View Project

Artemis

GenAI women safety application with real-time threat detection at 96% accuracy and Graph RAG navigation generating routes with 85% fewer high-risk areas. Now an active funded startup.

LangGraph React Native WebRTC Whisper

View Project

TransformoDocs

Machine-readable document converter with HyDE-based RAG pipelines; achieved 70% storage reduction and 36% conversion efficiency improvement via MLOps and agentic workflow optimization.

CrewAI MLFlow LangChain Streamlit

View Project

MoleculeForge

End-to-end molecular generation pipeline using fine-tuned Gemini models; reduced drug discovery time by 58% and achieved 76% higher validation success rate with self-reflective RAG on the ZINC dataset.

Gemini VAE GCNs RDKit AutoGen

View Project

NeuraTradePro

Sophisticated algorithmic trading platform leveraging neural networks and diverse technical indicators for cryptocurrency and financial market decision-making — end-to-end full stack application.

Neural Networks React Python Trading APIs

View All on GitHub

My Arsenal

Technical Skills

Languages

Python C/C++ JavaScript TypeScript Java SQL HTML5 CSS3

AI & ML

PyTorch TensorFlow LLM Optimization LoRA / QLoRA Hugging Face LangChain CrewAI Agentic RAG OpenCV LangSmith

Data Eng & Cloud

Apache Spark BigQuery GCP ETL Pipelines MLOps Docker AWS SageMaker MCP Servers Azkaban

Dev & Tools

React Node.js FastAPI Flask MongoDB WebRTC PowerBI Git

Beyond the Code

Leadership & Impact

Team Lead

Smart India Hackathon · 2023, 2024, 2025

Led multi-disciplinary teams for 3 consecutive years developing AI-driven solutions for national challenges — overseeing technical architecture, rapid prototyping, and final deployment.

Founder & CEO

Artemis Pvt Ltd · Startup Operations

Directing a 12-member team to build GenAI safety solutions; managing the full product lifecycle and investor relations while scaling the MVP into a production-ready business model.

Technical Head

E-CELL · Campus Entrepreneurship Cell

Leading a 20+ member tech team to automate summit registrations and optimize web infrastructure; improved digital engagement and turnout for national-level flagship events.

Get in Touch

Let's Build Something Together

Open to research collaborations, ML engineering roles, and startup partnerships. Drop me a message — I'd love to hear from you.