Debopam Chowdhury | ML & Software Engineer

About Me

Machine Learning & Software Engineer with expertise in ML/DL, MLOps, Mathematics, RAG, and Flutter. I'm passionate about building deployable AI-based solutions and have a strong foundation in Machine Learning, Computer Science, and Mathematics.

Currently working on LLM Pre-Training from scratch using PyTorch & Math, and recently completed a contract as a Generative AI Engineer. I've solved 100+ problems on LeetCode and pursuing my BE in Information Science from Acharya Institute of Technology (2021-2025).

Technical Skills

Programming Languages & Tools

Python Java Dart C++ Html-CSS-JS Apache Kafka Redis Git DBMS & ORM

Machine Learning & Deep Learning

TensorFlow scikit-learn Keras PyTorch Hugging Face Neural Networks LSTMs CNNs Transformers LLM Finetuning

Generative AI & RAG

Vector Embeddings Indexing Chunking RAG Pipelines LlamaIndex LangChain Colpali Byaldi Vector Databases LangGraph CrewAI

Cloud - Backend - DevOps

Docker AWS Hugging Face Spaces FastAPI CI/CD Pipeline

Mobile & Web Development

Flutter Firebase Riverpod Supabase Google OAuth

Experience

Apr,2025-Now

TurboML - SF,California

SWE-AI Intern - remote

Technologies:NLP, FastAPI, C++, Rust, Kafka, Agentic AI, gRPC

Sep,2024-Now

AI GYM BUDDY Side Hustle

Solo Creator: Design - Code - Deploy - Marketing

Personalized AI-driven workout app with smart equipment detection and progress tracking
650+ registered users
AI instrument detection (camera or gallery)
Personalized workout routines based on available equipment
Dynamic video tutorial finder
Google OAuth integration

Technologies: Dart, Flutter, Firebase, Gemini 2.0, Riverpod, LangChain, FastAPI, Google OAuth, Deep Learning

Google Play Store Website Demo Video

Oct-Nov 2024

Private Client, Sydney, Australia

Generative AI Engineer (Contract - Remote)

Developed secure, on-premise solutions for complex PDF with images and charts Q&A and knowledge retrieval
Built data ingestion pipeline for 1000's of documents with automatic task scheduler
Implemented multimodal RAG pipelines (Byaldi, Colqwen2, Pixtral 12B) optimized for diverse document types (70% accuracy improvement)
Containerized the application with Docker for deployment flexibility

Technologies: NLP, Vision Embeddings, Local Multimodal RAG, LangChain, Pixtral 12B, Col-Qwen2, Byaldi

Open-source contribution - Byaldi - 575 ✰

Nov-Dec 2024

Focus-flow, Remote

Flutter Developer (Contract)

Jan-June 2023

UNFILTR, INC, Bengaluru, India

Flutter Developer Intern

Additional Projects

Synthetic Data Generation

TurboML knowledge distillation Synthetic Data Generation - Q&A dataset about TurboML. Contains 1,343 technical questions with detailed answers covering implementation, troubleshooting, architecture design, and performance optimization.

Huggingface Dataset

Ongoing Projects

LLM - Continued Pre-Training (CPT + SFT) - 80% complete

Teaching an already existing Language Model (Qwen2.5-0.5B) to learn Bengali Language (Which it did not have knowledge about) using CPT (Continued Pre-Training) & then adapting it to, English to Bengali Translation, using SFT (Supervised FineTuning), all under 400MB memory for Edge Device Translation tasks. Using unsloth.

LLM-Pre-Training from Scratch - using Pytorch & Math

Pre-training a small LLama style foundational Model from scratch by only using Pytorch, Attention Libraries (Flash-Attention) and Math under 50 Dollars using Runpod-GPUs. Currently studying all Mathematical Concepts -- Plan to finish this project within 7 April, 2025.

Mini Projects

Efficient Parallel Implementation of Forward Prop, K-means using Math & Numpy Broadcasting.
Machine translation using Encoder-Decoder Architecture using Bi-directional LSTMs and Bahdanau Attention.
Math behind Gating mechanism of a LSTM & GRU cell.
Learning LangGraph & CrewAi.
Real-time Data-Center Anomaly Detection in Streaming Data with TurboML (HST + AdaBoost).

Volunteering

Git/GitHub Instructor

Taught juniors version control fundamentals

Event Coordinator

Acharya Technical Club - Steigen

Non-Technical

Video Editing

Utkarsh Cultural Club (2022)
The Boring Edit Agency

Hackathon

Amazon ML Challenge

Ranked 172/75,000 - Image Entity Extraction

Processed 71,000/131,000 images (54%) then Ran out of time
Made large vision entity extractor pipeline with VLM (Qwen2VL) & post formatting via Regex

Education

2021-2025

BE in Information Science

Acharya Institute of Technology, Bangalore

CGPA: 8.32

2019-2021

Higher Secondary Education

Kalyani Public School, Barasat, Kolkata

77% (Covid Batch - No Exam - AutoPass)

2019

Secondary Education

Sacred Heart Day High School, Kolkata

90%

Hi, I'm Debopam Chowdhury