AI / ML Engineer Open to opportunities

Software engineer · AI / ML engineer

Software engineerworking withvision & language.

I’m Shoumik. I build computer vision and multimodal models, plus the APIs and web applications that use them.

Explore my work Get in touch
GitHub ↗LinkedIn ↗Email ↗
Portrait of Shoumik Sarkar
Looking forSWE & AI/ML roles
Dhaka, BDLocal time · Dhaka
Scroll to explore ↓
01 Computer vision02 Multimodal AI03 Product systems
Computer vision, from data preparation to deployment.
02 / ABOUT

About Shoumik

Working with
vision and language.

My projects combine images, text, and audio for deception detection, misinformation classification, and search. I work on data preparation, model evaluation, and deployment.

Based inDhaka, Bangladesh
Available forOpen to internships & full-time roles.

What I do

Computer Vision

Extracting facial landmarks and image features with MediaPipe, OpenCV, and ResNet-18.

Language & Multimodal AI

Combining OCR text, image features, and language models to classify social media posts.

Audio Intelligence

Extracting audio features with Librosa and combining them with facial and visual features.

AI Product Systems

Building FastAPI services and web interfaces for model inference, search, and user accounts.

My approach

  1. 01

    Understand

    Define the task.
    Inspect the data.

  2. 02

    Build

    Start with a baseline.
    Test model changes.

  3. 03

    Validate

    Measure performance.
    Check failure cases.

Use the results to plan the next experiment.

Recent experience

Machine Learning Engineer Intern

Flyrank AI

JUN - SEP 2026

Worked on ML pipelines, dataset evaluation, Python analysis, and deployment with the engineering team.

Applied MLData pipelinesProduct development
03 / SELECTED WORK

Projects &
research.

University research projects, a football trivia site, and a research collaboration app in development.

DeceptiLens website homepage previewExplore live site ↗
01 / Senior Design ProjectLive project

DeceptiLens

A deception-detection model combining facial landmarks, audio features, and visual embeddings through cross-attention. Includes a Streamlit app for testing video samples.

PyTorchMediaPipeLibrosaStreamlit
Project notes
  • Built an end-to-end pipeline combining facial landmarks, audio descriptors, and deep visual embeddings.
  • Fused MediaPipe, Librosa, and ResNet-18 features with a custom cross-attention model.
  • Achieved 82.11% accuracy on the DOLOS benchmark.
  • Deployed an interactive Streamlit app with real-time performance visualisations.
Live site GitHub
Football Genius website homepage previewExplore live site ↗
02 / Full-Stack Football Trivia PlatformLive project

Football Genius

A football trivia site with multiple game modes, player statistics, accounts, and leaderboards. Built with JavaScript, Supabase, and the Gemini API.

JavaScriptSupabaseGemini APIGitHub Actions
Project notes
  • Built and deployed a mobile-first, multi-game-mode trivia platform with a custom dark-mode UI.
  • Added email, Google OAuth, and OTP authentication, plus dashboards, scores, and profiles.
  • Automated player-stat updates with a scheduled GitHub Actions Node.js scraper and Supabase sync.
  • Connected Supabase authentication and storage with Google sign-in and the Gemini API.
Live site GitHub
Team Scholaris website homepage previewDevelopment preview ↗
03 / Research Collaboration PlatformIn development

Team Scholaris

A research collaboration app for shared tasks, discussions, files, and paper writing. Still in development and not yet open for public use.

Research workflowsCollaborationProduct development
Project notes
  • Organizes research tasks and paper-writing work in a shared workspace.
  • Keeps team discussions and research files alongside writing tasks.
  • Under active development; the source repository remains private.
Preview site Private repository
04 / Directed Research Project2025

RumorFusionX

A misinformation classifier for multilingual social media posts. Combines image features, OCR text, and descriptions generated by a language model.

PyTorchOCRTransformersFastAPI
Project notes
  • Combined OCR-extracted text, image features, and LLM-generated contextual descriptions.
  • Collected, cleaned, and analysed a multilingual dataset of 2,500 social-media images.
  • Built REST-backed feature fusion services that achieved 93.07% classification accuracy over unimodal baselines.
GitHub
05 / Junior Design Project2024–25

MedHelp

A medical information chatbot that retrieves reference material to answer questions. Also collects medicine prices for comparison.

TransformersFastAPIPostgreSQLRAG
Project notes
  • Helped scope and build a conversational AI system using retrieval-augmented generation.
  • Processed structured medical reference data to improve response relevance.
  • Implemented medicine-price aggregation for comparative search.
GitHub
04 / SKILLS

Tools I use.

Select a tool to see how I use it and my experience level.

Core AI/ML Expertise

Deep Learning

Training neural networks, extracting features, and comparing model performance.

NLP & LLMs

Using Transformers and retrieval to answer questions with reference material.

Computer Vision

Preparing images, extracting facial landmarks, and computing visual embeddings.

Multimodal Learning

Combining image, audio, and text features using cross-attention.

Frameworks & Libraries

Languages

Tools & Platforms

Experience LevelChoose a node
BasicIntermediateAdvanced
05 / EXPERIENCE

Engineering
experience.

JUN - SEP 2026

Machine Learning
Engineer (Intern)

Flyrank AI

Worked with senior engineers on ML pipelines and deployment. Evaluated datasets, analysed model performance in Python, and helped turn system requirements into implementation tasks.

MAY - SEP 2026

Research
Engineer

Verdant

Researched client requirements and wrote technical specifications before development. Worked on API integration and database management, and helped the team define scope and plan deliverables.

06 / EDUCATION

Education &
training.

01 / Degree

BSc, Computer Science & Engineering

North South University · Graduated May 2026

Relevant Coursework

Data Structures & AlgorithmsTheory of ComputationProgramming Languages (C, C++, Java)Machine LearningDeep LearningPattern Recognition & Neural NetworksComputer Vision
02 / Certification

Vision Language Models Bootcamp

Vision Language Models Bootcamp certificate awarded to Shoumik Sarkar

OpenCV University, 2026. Completed the Vision Language Models Bootcamp with a grade of 92%.