Hi, I'm
HARSHIT DHAR
AI/ML Engineering Undergraduate building scalable intelligence, deep learning systems, and data-driven solutions.
About
I am a Computer Science undergraduate specializing in Artificial Intelligence and Machine Learning. With a strong foundation in scalable software architectures and applied mathematics, my primary focus is developing high-performance intelligent systems that solve complex, real-world problems.
My expertise spans modern deep learning frameworks like TensorFlow and PyTorch, alongside robust backend development in Python. I thrive in dynamic, fast-paced environments where rigorous engineering standards and rapid iteration intersect, having proven this through successful outcomes in numerous international hackathons and active contributions to open-source software.
I am actively seeking internship opportunities where I can contribute to mission-critical initiatives, scale machine learning pipelines, and further bridge the gap between algorithmic research and production-grade software.
Selected Work
A curated selection of projects showcasing my problem-solving approach and technical depth.
Intent Forge — Nova AI Writing Assistant
Generic AI writing tools lack control over tone, audience, and content purpose.
Developed a full-stack AI application that generates structured content using intent-driven prompt engineering. Supports multiple formats (blogs, emails, pitches) with tone and audience customization, along with speech output.
Improved content relevance and usability by enabling tailored, human-like text generation. Deployed scalable system using Vercel and Render.
Ferwine — AI-Powered Discord Assistant
High message volume in Discord servers reduces readability and limits meaningful engagement.
Developed a Discord bot using discord.py integrated with LLM APIs to perform conversation summarization, sentiment analysis, topic extraction, and context-aware prompt generation. Built a tweet-generation system for converting messages into concise formats and implemented data storage using PostgreSQL.
Enabled real-time insights and improved engagement in community discussions by transforming raw chat data into structured, actionable information. Deployed scalable bot infrastructure using Railway.
MediX — Medical Text Simplification
Clinical text is often inaccessible to non-experts, limiting patient understanding.
Developed a transformer-based simplification pipeline using T5-small and T5-base trained on the Med-EASi dataset. Evaluated model performance using readability metrics (FKGL, ARI) and implemented cascaded inference with FLAN-T5 for improved output quality.
Achieved ~2 grade-level reduction in readability while maintaining clinical accuracy. Showed effectiveness of combining domain-specific fine-tuning with instruction-tuned models for real-world text simplification.
Lung Disease Prediction with Explainable AI
Traditional ML models for medical prediction lack interpretability, limiting their usability in real-world healthcare settings.
Developed a classification pipeline using Logistic Regression with standardized features and cross-validation. Applied SHAP for feature attribution and designed a fuzzy logic system to map probability outputs into interpretable risk levels (Low, Medium, High).
Enabled transparent decision-making by combining model predictions with human-readable explanations, improving trust and interpretability in medical risk assessment systems.
Kohabit — Roommate & Friend Recommendation
Traditional roommate matching relies on basic filters and fails to capture deeper compatibility between individuals.
Developed a vector-based recommendation system by transforming user inputs (preferences, interests, lifestyle data) into embeddings. Stored vectors using pgvector and performed similarity search using cosine similarity to identify semantically similar profiles.
Improved match quality by leveraging embedding-based retrieval instead of rule-based filtering. Built a scalable pipeline for real-time recommendations and deployed using Vercel and Render.
Hair Tracker AI
Tracking dermatological progress requires subjective visual comparison, making treatment efficacy hard to measure.
Engineered a computer vision application utilizing OpenCV and TensorFlow, deployed interactively via Streamlit.
Provided users with an automated, objective tracking system for monitoring hair growth and scalp health over time.
ASL-VISION
Lack of accessible real-time translation tools limits communication for individuals using American Sign Language.
Developed a robust computer vision pipeline using PyTorch to classify static ASL gestures from diverse pictorial sources.
Established a scalable deep learning architecture capable of real-time sign language recognition with high accuracy.
Music Genre Classification
Manual categorization of vast music libraries is highly inefficient and subjective.
Extracted audio features using Librosa and engineered an RNN-based sequence model trained on the GTZAN dataset.
Automated genre tagging with robust predictive performance across diverse musical categories.
Experience & Involvements
Open Source Contributor
Hacktoberfest 2025
- Actively contributed to multiple cutting-edge open-source repositories.
- Resolved complex bugs and implemented technical enhancements reviewed by maintainers.
- Gained hands-on experience in large-scale codebase navigation and version control.
Core Member (Research Team)
Artificial Intelligence Society Bennett
- Contributed to the AI Research Team, focusing on advanced Machine Learning, Networking, and large-scale applications.
- Collaborated with peers to dissect, analyze, and implement state-of-the-art AI research papers.
- Engaged in hands-on workshops and continuous skill development spanning over 9+ core technical areas.
Freelance Developer
Independent Contractor
- Developed web applications and AI-integrated solutions for clients
- Managed full development lifecycle from requirements to deployment
- Delivered responsive, user-focused interfaces
Hackathon Competitor
Various Global Hackathons
- Participated in 7+ competitive hackathons focusing on AI and scalable software solutions.
- Collaborated in agile, high-pressure environments to ship viable MVPs within 24-48 hour constraints.
- Pitched technical architectures and product roadmaps to senior industry judges.
Writing & Insights
How I fine-tuned T5 to simplify complex medical texts
A deep dive into training the T5 transformer model to translate dense medical jargon into accessible language for non-experts.
My Journey into AI
Reflections on my transition into Artificial Intelligence, the mathematical foundations I focused on, and the critical paradigm shifts.
Technical Stack
Machine Learning & AI
Core Computer Science
Languages
Tools & Infrastructure
Professional Skills
Certifications & Badges
Verified Badges
Academic & Specialized Coursework
AI Fluency for Students
Claude 101
Certified System Administrator (CSA)
Retrieval Augmented Generation (RAG)
Accelerated Computing in Modern CUDA C++
Convolutional Neural Networks in TensorFlow
Algorithmic Toolbox
Data Analysis with R Programming
Introduction to Bayesian Statistics
Programming with Python
Data Structures
Intro to Machine Learning
Intro to AI Ethics
Object-Oriented Data Structures in C++
Career Essentials in Sustainable Tech
Introduction To Generative AI
AI For Everyone
Get in Touch
Let's build something.
Whether you have an internship opportunity, a project proposal, or just want to connect — my inbox is always open.