BERT, from scratch
A 7.5M-parameter Bengali BERT, pre-trained from nothing on a laptop, that scores above published mBERT and IndicBERT on Bengali news classification.
Samyamoy Rakshit
Data Scientist
I rebuild published models from the paper — the transformer, BERT, ViT — tensor by tensor in PyTorch, all three on Bengali data, all three trained end to end on one M1 laptop with 16 GB of RAM. Reading a paper and reimplementing it turn out to be very different kinds of understanding. By day I build forecasting and agentic systems at Sigmoid.
Bengaluru, India
Experience
- Jan 2026 – present
Data Scientist · Sigmoid
Re-engineered daily demand forecasting across 9,200 machine × product series — full run from ~3 hours to ~3 minutes. Plus an MCP proof of concept driven from three model backends.
- Dec 2024 – Dec 2025
Associate Data Scientist · Sigmoid
Built the feature-engineering pipeline behind a four-agent RAG recommendation engine, and worked on one of those agents — live since June 2025, ~80% accuracy in its core categories. Earlier, price elasticity and promo baselines for a global CPG manufacturer.
- Aug 2024 – Dec 2024
Data Science Intern · Sigmoid
Measured which levers — price, promo, media, retail execution — actually move sell-through for a global spirits manufacturer, so spend could be aimed where it pays.
Projects
papers-from-scratchSOTA architectures rebuilt in PyTorch
Three papers replicated tensor by tensor on Bengali data — no nn.Transformer, no nn.MultiheadAttention, no pre-built HuggingFace models.
BERT
BERT: Pre-training of Deep Bidirectional Transformers for Language UnderstandingDevlin et al., 2019
A 7.5M-parameter Bengali BERT, pre-trained from nothing on a laptop, that scores above published mBERT and IndicBERT on Bengali news classification.
- Params
- ~7.5M
- Corpus
- ~114MB
- Pre-trained
- ~28h
- Accuracy
- 86.5%
Transformer
Attention Is All You NeedVaswani et al., 2017
The original transformer rebuilt component by component, then trained to translate English into Bengali on one laptop.
- Params
- ~11M
- Paper
- ~65M
- Hardware
- M1, 16GB
- Data
- Samanantar
Vision Transformer
An Image Is Worth 16x16 WordsDosovitskiy et al., 2021
A ViT trained on 1,935 photographs of Bengali terracotta temples — a dataset I had to build first — to test the paper's claim about pre-training. It failed, exactly as predicted.
- Images
- 1,935
- Classes
- 10
- From scratch
- 14.9%
- Pre-trained ViT-B/16
- 88.9%
Education
- Nov 2020 – Jul 2024
B.Tech, Computer Science and Engineering (Data Science)
Haldia Institute of Technology
CGPA 9.05. Affiliated to Maulana Abul Kalam Azad University of Technology, West Bengal.
- 2020
Higher Secondary (WBCHSE), Science — 94.4%
Jhantipahari High School
- 2018
Secondary (WBBSE) — 90.4%
Jhantipahari High School
Writing
Attention Is All You Need, from scratch
The original transformer rebuilt component by component, then trained to translate English into Bengali on one laptop.
Vision Transformer, from scratch
A ViT trained on 1,935 photographs of Bengali terracotta temples — a dataset I had to build first — to test the paper's claim about pre-training. It failed, exactly as predicted.
Skills
Everything here is something I have shipped with or trained with.
Languages
- Python
- SQL
Deep learning
- PyTorch
- Transformers
- BERT
- Vision Transformers
Generative and agentic
- Azure OpenAI
- Azure ML Prompt Flow
- RAG
- Multi-agent systems
- Model Context Protocol
- Ollama
Modelling and statistics
- scikit-learn
- LightGBM
- Prophet
- Mixed-effects models
- SARIMAX
- Ridge and Lasso
- Experiment design
Data and cloud
- Snowflake
- Microsoft Fabric
- Azure ML