Koshik Debanath kowshik24

Hi, I'm Koshik 👋

📍 Bangladesh ↔ Remote | 🧠 GenAI Researcher | 🛠️ Full Stack ML Engineer

Bridging the gap between complex reasoning research and production-grade AI. Building novel multi-modal models and shipping open-source tools.

🔭 Portfolio – My research, projects, and journey.

⚙️ ConfigSync - A CLI that treats configuration as code, with versioning, diffing, and team-wide enforcement.
🏠 HomeOps - An iOS app that is essentially a "Digital Glovebox" for the home.
⚡ FOMI - Frequency-Optimized Manifold Indexing. High-performance library for image vector similarity search & clustering.
🔄 Atomic Sync - Real-time, end-to-end encrypted sync for Obsidian using Supabase.
📄 PaperToCode - Converts research papers into executable code implementations using GenAI.
📹 Loom Clone - Cloud-native video messaging platform with direct Google Drive storage integration.
🕵️ SynthDetect Ultra - Forensic analysis tool distinguishing authentic photos from AI-generated media (GANs/Diffusion).
🧠 Obsidian AI Plugin - Context-aware AI assistant with multi-file context and streaming, built for Obsidian users.
💹 Binance AI Agent - Multi-agent system providing crypto investment recommendations via OpenAI/Gemini APIs.
🇧🇩 Bengali Semantic Retrieval - Optimizing retrieval for low-resource languages using Matryoshka Representation Learning.
📉 Stock Forecast BD - LSTM models forecasting stock prices for Bangladeshi and global markets.
🧩 PineconeUtils - Authored two Python libraries to simplify data handling for RAG systems.

🛠️ OpenLLMetry - Fixed serialization bugs for Python dataclasses and TypeErrors in OpenAI embeddings metrics (PR #2800, #1836).
🌲 Pinecone Canopy - Contributed to the Retrieval-Augmented Generation (RAG) framework.

[Mathematical Biosciences] Bayesian Physics-Informed Neural Networks for Parameter Inference in Wound Healing (Under Review, 2025)
[TALLIP] Optimizing Semantic Retrieval for Bengali: A Comparative Analysis (Under Review)
[ICCIT 2023] An Attention-Based Deep Learning Approach to Knee Injury Classification from MRI Images
[ECCE 2025] Advancing Low-Resource NLP: Contextual Question Answering for Bengali Language Using Llama
[NCIM 2025] Distinguishing Human-Written and AI-Generated Text: A Comprehensive Study Using XAI

Random Facts

🧠 Research focus: Explainable AI (XAI), Low-Resource NLP, Large Languge Models(LLMs) and ML/DL.
💻 Expert in Python, but currently building decentralized governance platforms with Solidity & React.
📚 Researcher at Young Learners' Research Lab (YLRL).
⚡ Focused on optimizing LLMs via Matryoshka Representation Learning.