Devansh Kumar - Portfolio

About Me

AI Developer and Research Engineer with experience in building and optimizing Large Language Model (LLM) systems, AI-powered platforms, and scalable machine learning pipelines. Proven ability to improve performance metrics significantly through fine-tuning, prompt engineering, and efficient system design. Currently pursuing a Master's in Computer Science with a concentration in Applied AI at the University of Ottawa. [cite: 1, 2, 6, 10, 21]

Ottawa, Ontario (343) 777-9132 devansh.kumar.kd@gmail.com [cite: 1]

Experience

AI Developer Intern

Accenture R&D and uOttawa | Jan 2025 - Apr 2025

Contributing to the SALI data insights platform, enhancing query understanding, retrieval accuracy, and scalability. [cite: 1]

AI Research Engineer Intern

Huawei Technologies | May 2024 - Dec 2024

Developed PoC Agentic AI LLM systems (LangChain, LangGraph, Transformers, PyTorch) for factory automation. [cite: 2]
Improved inference performance by 2x via process optimization and prompt engineering. [cite: 3]
Reduced local LLM inference time by 95% using parameter tuning and advanced prompt techniques (CoT, Self-Consistency). [cite: 4]
Contributed to IEEE GLOBECOM'24 presentation on Multi-Agent LLM systems. [cite: 5]

AI Developer Intern

Accenture R&D and uOttawa | Jan 2024 - Apr 2024

Implemented rerankers & HyDE methodology with LLMs to boost knowledge graph retrieval and robustness. [cite: 6]
Fine-tuned embedding models, achieving 13% relevance improvement on Spyder benchmark. [cite: 7]
Designed proprietary evaluation metrics for RAG framework analysis. [cite: 8]
Utilized Azure OpenAI; contributed to CI/CD pipelines with comprehensive test cases. [cite: 9]

Software Developer PRISM Intern

Samsung R&D Institute | Nov 2021 - Oct 2022

Achieved 96% accuracy for Bengali entity recognition/intent classification by finetuning BERT & Spacy NLU models. [cite: 10]
Reduced Jarvis ASR model latency by 10% for faster voice recognition. [cite: 11]
Integrated Spacy and Jarvis models with Rasa for Bengali NLP customer service. [cite: 12]

Projects

AI-Powered Grocery Optimization Platform

Built a full-stack platform (React, Tailwind, Node.js, FastAPI) for real-time price tracking & meal planning. [cite: 13]
Developed LangGraph AI agents for price/nutrition analysis and meal recommendations. [cite: 14]
Designed PostgreSQL DB (10k+ items) with optimized queries. [cite: 15]
Automated real-time price scraping using Selenium. [cite: 16]
Integrated AI decision layers for cost/nutrition optimization. [cite: 17]

StockPro MLFlow: Scalable AI Pipeline

Built scalable ML pipelines (ZenML) reducing workflow time by 30%. [cite: 18]
Deployed LSTM models (TensorFlow/Keras) achieving 85% accuracy; used MLflow for versioning/tracking. [cite: 19]
Enhanced model performance by 12% using Random Forest feature selection & MinMax scaling; deployed low-latency endpoints via MLflow. [cite: 20]

Skills

Programming Languages

Python
C++
R
Java
Matlab
JavaScript
TypeScript

[cite: 21]

LLM Frameworks

LangChain
LangGraph
Ollama
Autogen

[cite: 21]

Databases & Vector Search

PostgreSQL
MySQL
MongoDB
NoSQL
FAISS
ChromaDB
OpenSearch

[cite: 21]

ML & AI Libraries

PyTorch
TensorFlow
spaCy
NLTK
scikit-learn
Hugging Face Transformers
MLFlow
Weights & Biases
Kubeflow
Pandas
NumPy
Matplotlib

[cite: 21]

Software Development

Docker
Git
GitHub
Jira
CI/CD Pipelines

[cite: 21]

Cloud Computing

AWS
Azure

[cite: 21]

Web Technologies

React
Node.js
FastAPI
Express.js
Selenium

[cite: 21]

DevOps & Containerization

Kubernetes
CI/CD Automation

[cite: 21]

Education

Master of Computer Science & Concentration in Applied Artificial Intelligence

University of Ottawa, Ottawa, Canada

Sept 2023 - Graduation: April 2025

Anticipated CGPA: 9.57/10.00

[cite: 21]

B.Tech. in Computer Science and Engineering

Vellore Institute of Technology, Chennai, India

July 2019 - May 2023

CGPA: 8.72/10.00

[cite: 22]