AI/ML • Data Science • Math Olympiad

Hi I'm Prachya Biswas

Computer Science student at Brac University

Trying to learn what machines learn

Regional Champion at BdMO · Problem setter at Bangladesh Mathematical Olympiad

Portrait of Prachya Biswas
Prachya Biswas
Dhaka, Bangladesh

Experience

ML & AI Research Lab, Bangladesh

Research Secretary

Apr 7, 2026 – Present · 7 months

Dhaka, Bangladesh[tag]

Secretary of a research lab for machine learning and AI in Bangladesh.

Bangladesh Mathematical Olympiad

Academic Team Member

Dec 11, 2025 – Present · 11 months

Dhaka, Bangladesh[tag]

Academic team member at the national math olympiad.

IEEE BRAC University Student Branch

Secretariat

Jun 25, 2026 – Present · 5 months

Dhaka, Bangladesh[tag]

Part of the branch secretariat.

Honors and achievements

Honors and achievements

Participant, International Data Science Olympiad (IDSOL)

2026

Presented Vangari AI and led the data science side of the team.

Former Regional Champion, Bangladesh Mathematical Olympiad

Champion at the regional round of BdMO.

IBM Machine Learning Professional Certificate

Professional certificate from the IBM Skills Network.

National Children's Award

2019

First place in both the Quiz Contest and the Science Project Contest.

Former Promising Debater

Recognised at the RC CTG.mid CITY-Drishti School of Debate.

Competitions entered

International Machine Learning Olympiad (IMLO)

2026

Team of 3.

NASA Space Apps Challenge, Dhaka

2026

Data Science Lead, team Agnishikha, project AstraFlame.

Skills

Languages

  • Python
  • Java
  • TypeScript

Machine Learning

  • Linear Regression
  • Logistic Regression
  • KNN
  • SVM
  • Neural Networks
  • Time Series Analysis
  • Gaussian Processes

Data Science

  • Scikit-Learn
  • NumPy
  • Pandas
  • SciPy
  • Matplotlib
  • Jupyter Notebook
  • Data Preprocessing & Feature Engineering
  • Data Pipelines
  • Model Evaluation & Cross-Validation

AI

  • RAG Pipelines
  • AI Agents
  • Agentic AI
  • LangChain

Deep Learning

  • PyTorch
  • timm
  • TensorFlow
  • Vision Transformer (ViT)
  • EfficientNet

Backend and Web

  • Next.js
  • React
  • FastAPI
  • Express.js

Databases

  • PostgreSQL
  • PostGIS

Tools

  • Git
  • GitHub
  • Google Colab

DevOps and Infrastructure

  • Docker
  • Docker Compose
  • GitHub Actions
  • Linux
  • Azure

Mathematics

  • Olympiad Problem Solving
  • Teaching

Spoken Languages

  • Bangla
  • English

Projects

MediSense BD

An integrated national health intelligence platform for Bangladesh. It forecasts dengue, diarrhea, influenza, typhoid and malaria risk across all 64 districts, and adds hospital routing, drug verification and Bengali symptom triage. Prachya implemented the TRACE forecasting model as data scientist.

  • Python
  • scikit-learn
  • LSTM
  • Next.js
  • PostgreSQL

Vangari AI

A waste marketplace that connects waste owners with certified collectors. A photo of scrap goes in, and the AI estimates the material mix, weight and payout, then the nearest collector is dispatched. Built for IDSOL 2026, with Prachya leading the data science side.

  • Gemma 4
  • PyTorch
  • ViT-Small
  • EfficientNet-B0
  • Next.js

AstraFlame

Team AgniShikha's entry for NASA Space Apps Challenge 2026 (Dhaka), "Flame in Freefall". From oxygen level, pressure, airflow and material it estimates flame spread, extinction and smoke, trained on 850 NASA microgravity combustion runs, with error bands and a warning when a condition was never tested. Prachya was Data Science Lead.

  • Python
  • scikit-learn
  • Gaussian process
  • Next.js
  • NASA data

Agentic-AI-From-Scratch

A beginner-friendly learning repo on generative AI, RAG and agentic AI with LangChain, built module by module. It has theory notes and working code: first LLM calls, a blood report analysis app with a Streamlit UI, ChromaDB semantic search and a RAG pipeline.

  • Python
  • LangChain
  • RAG
  • ChromaDB
  • Streamlit

Machine learning algorithms from scratch

Linear regression, logistic regression, KNN, SVM and a neural network, each written from scratch in NumPy without ML libraries.

  • Python
  • NumPy

Computational Biology Meets Cells

Module-by-module notes from self-study of bioinformatics, with annotated commands and solved exercises. The 17-module plan covers sequence alignment, genome assembly, protein structure, phylogenetics, RNA-seq and multi-omics, with single-cell analysis (Seurat, CellChat, Monocle) planned.

  • Bioinformatics
  • Python
  • R
  • Biopython
  • Conda

Leadership and Mentorship

Math olympiad academic team

Academic team member at the Bangladesh Mathematical Olympiad.

Teaching olympiad students

Prep classes for Prothom Alo Math Olympiad students, from primary to advanced level. The first class was a two-hour lecture on the Pigeonhole Principle.

Study materials

Worksheets and exam guides for Cambridge O Level Mathematics students.

Seminars and workshops

Sessions at CUET and Chittagong University (details under Talks and Features).

Student communities

Secretariat at IEEE BRAC University Student Branch, Sub-Executive of HR at BUEDF, General Member in R&D at BUCC.

Writing

Notes on machine learning, experiments and the things I learn along the way.

Featured essaySample article

Evaluating Machine Learning Models Beyond Accuracy

October 12, 20262 min read

Accuracy is a starting point, not a verdict. Here is a practical checklist for choosing metrics, building honest splits and doing error analysis before you ship a model.

  • #Machine Learning
  • #Evaluation
  • #Python
Read the essay

More from the journal

1 article

More essays are on the way.

Contact Me

Have a project, a research idea or a question about math olympiads? Send me a message.