Deep Learning & Computer Vision Engineer

Mrinal Bharadwaj

******.**********@*****.*** • +1-832-***-**** • OR, USA • LinkedIn • GitHub

SUMMARY

Computer Science graduate (M.Eng, Oregon State University) focused on deep learning and computer vision. Hands-on experience training CNN, ResNet, and transformer models in PyTorch, and improving model reliability through careful data analysis and evaluation. Seeking an entry-level Machine Learning or Computer Vision Engineer role. EDUCATION

Oregon State University — M.Eng, Computer Science Sep 2024 – Dec 2026 Relevant coursework: Machine Learning, Deep Learning, Natural Language Processing, Artificial Intelligence, Computer Vision. Noida International University — B.Tech, Computer Science May 2018 –May 2022 TECHNICAL SKILLS

Languages Python, C, C++

ML / Deep Learning PyTorch, TensorFlow, CNNs, ResNet, Transformers, Domain Adaptation Computer Vision Image classification, 3D reconstruction, real-time detection Tools & Platforms CUDA / GPU training workflows, Git, Linux, NumPy, pandas EXPERIENCE

Machine Learning Research Intern Jun 2025 – Present Oregon State University

- Trained and evaluated CNN and ResNet-18 image models in PyTorch, using CUDA-enabled GPU workflows to accelerate experiments.

- Diagnosed a failing model by tracing the problem to data quality — cleaned null values, corrected data alignment, and selected the correct lighting dataset — substantially improving model performance.

- Added data validation checks to catch mislabeled and inconsistent samples before training. PROJECTS

Bridging the Synthetic-to-Real Gap in 3D Object Reconstruction Graduate Deep Learning

• Built a hybrid CNN–Transformer model (~17.5M parameters) to study the synthetic-to-real gap in single-image, feed- forward 3D reconstruction.

• Compared reconstruction architectures (Pix2Vox++, TripoSR, LGM) trained on synthetic Cap3D renders and evaluated on real photographs.

• Applied domain-adaptation techniques — test-time adaptation, DANN, AdaIN style transfer, and augmentation — to measure their effect on modern architectures.

The Battle of Neighborhoods — Data Science Capstone (IBM Applied Data Science)

• Collected data about Toronto neighborhoods — pulled location info from Wikipedia (using BeautifulSoup and Requests) and nearby-places info from the Foursquare API, then organized it all into tables with pandas.

• Used a grouping method called k-means (from scikit-learn) to sort 103 areas into 10 groups of similar neighborhoods, and compared them by house prices and school ratings.

• Made interactive maps with Folium so someone moving to a new city could easily see and pick a good neighborhood.

• it's a project that gathers neighborhood data, groups similar areas together, and shows them on a map to help people choose where to live.

RPG Text Adventure — Personal Project (C++)

• Built a turn-based role-playing game that runs in the console, written fully in C++, where the player battles enemies by choosing actions.

• Designed a level-up system (levels 1–6) where the player's health, damage, and defense grow, and new abilities unlock as they progress — like Critical Hit, Blocker, Life Steal, and Ranged Attack.

• Created enemies with their own stats and a final boss (Murloc) with special moves — summon a minion, heal itself, and hypnotize the player — using random damage ranges so fights aren't the same every time. Tools: C++, developed on Replit.

Real-Time Face Mask Detection System Computer Vision

- Developed a CNN-based system to detect face-mask usage in real-time video for COVID-19 public-safety screening.

- Combined Haar Cascade face detection with a convolutional classifier for live, frame-by-frame inference.

Contact this candidate