Naishadh Vytla
███╗ ██╗ █████╗ ██╗███████╗██╗ ██╗ █████╗ ██████╗ ██╗ ██╗ ████╗ ██║██╔══██╗██║██╔════╝██║ ██║██╔══██╗██╔══██╗██║ ██║ ██╔██╗ ██║███████║██║███████╗███████║███████║██║ ██║███████║ ██║╚██╗██║██╔══██║██║╚════██║██╔══██║██╔══██║██║ ██║██╔══██║ ██║ ╚████║██║ ██║██║███████║██║ ██║██║ ██║██████╔╝██║ ██║ ╚═╝ ╚═══╝╚═╝ ╚═╝╚═╝╚══════╝╚═╝ ╚═╝╚═╝ ╚═╝╚═════╝ ╚═╝ ╚═╝ ██╗ ██╗██╗ ██╗████████╗██╗ █████╗ ██║ ██║╚██╗ ██╔╝╚══██╔══╝██║ ██╔══██╗ ██║ ██║ ╚████╔╝ ██║ ██║ ███████║ ╚██╗ ██╔╝ ╚██╔╝ ██║ ██║ ██╔══██║ ╚████╔╝ ██║ ██║ ███████╗██║ ██║ ╚═══╝ ╚═╝ ╚═╝ ╚══════╝╚═╝ ╚═╝
About
USER........: naishadh vytla ROLE........: ai/ml ops engineer in training INSTITUTION.: kl university DEGREE......: b.tech, electronics & communication engineering STATUS......: pre-final year, shipping TIMEZONE....: asia/kolkata (UTC+5:30)
ABOUT
i'm a pre-final year engineer working at the edge of applied deep learning research. my focus is vision transformers for medical imaging — training architectures that are small enough to deploy, accurate enough to matter, and transparent enough to trust.
i also ship. privacy-first ml tools, retrieval-augmented systems, saas products that solve concrete problems. recently a finalist at the visa 24hrs ai hackathon at iit madras, currently deepening my work in computer vision and continual learning.
Experience
drwxr-xr-x naishadh 5.0K may 2025 — jul 2025 smart_swift_innovations/
data science intern
- built classification and regression models on real-time operational datasets in python, scikit-learn, pandas — reached 92% accuracy on the primary prediction task.
- ran exploratory data analysis across 5+ datasets; surfaced feature correlations that improved model performance by ~15%.
- automated data cleaning pipelines, cutting preprocessing time by 40% across recurring workflows.
drwxr-xr-x naishadh 2.1K aug 2025 — feb 2026 kl_sac_adventure_club/
co-lead, adventure club
- directed outdoor expeditions for 50+ members — end-to-end logistics, safety protocols, and team coordination across multiple events.
Projects
~/projects/ ├── BrainTumorViT/ [ACTIVE] vision transformer · medical imaging ├── ResumeScreener/ [SHIPPED] privacy-first on-device screener ├── RAGDocQA/ [SHIPPED] retrieval-augmented qa over docs └── VISAHackathon24h/ ◆ [FINALIST] iit madras · 2026
| status : [ACTIVE] ongoing research | | stack : pytorch · timm · vision transformer | | | | single vit-b/16 backbone classifying 4-class brain | | mri (glioma, meningioma, pituitary, no tumor) at | | 99.17% accuracy — outperforming multi-model | | ensembles on the nickparvar benchmark. three-phase | | training with discriminative learning rates, | | stochastic weight averaging, mixup (a=0.4) and | | cutmix (a=1.0), a custom mlp head (768->512->256->4)| | and gradcam-based explainability. | | | | $ cd BrainTumorViT && open README | | [ → GITHUB ] |
privacy-first automated resume screener running fully on-device. llama 3 via ollama performs semantic matching between job descriptions and candidate resumes with zero external data transfer. pdf parsing via pypdf2, cosine-similarity ranking, clean streamlit ui.
[ → GITHUB ]retrieval-augmented qa system for pdf and docx files. langchain orchestration, faiss vector store, sub-second retrieval. users upload documents and query them in natural language, answers grounded in retrieved context — no hallucinated citations.
[ → GITHUB ]built and deployed a live ai-powered web application in 24 hours at iit madras. advanced to the finals among competing teams from across the country.
[ → DEMO ]Arcade
> a terminal-native homage to the chrome offline dino. jump cacti, don't die. hi-score persists via localStorage.
Skills
.--------. / __ __ \ | |_/ \_| | | ______ | | | | | | |______| | \__________/ ||| |||