Skip to content
View vp916's full-sized avatar

Highlights

  • Pro

Block or report vp916

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
vp916/README.md

Terminal: python train.py --model varun-prajapati --epochs ∞. CSE at MSU Baroda plus BS Data Science at IIT Madras. Still training.

LinkedIn Email Profile views

class Varun(Engineer):
    loss_fn   = "the gap between a notebook and a shipped product"
    optimizer = "read the paper → break the baseline → ship the API"
    weights   = {"deep_learning": 0.45, "classical_ml": 0.25, "full_stack": 0.30}

    def forward(self, problem):
        model = self.train(problem, from_scratch=True)   # u-net++ dice 0.9796, no pretrained weights
        api   = FastAPI(model)                            # served as a microservice
        return React(api)                                 # …and someone actually clicks it

⟢ model.summary()

Stack drawn as a neural network. Input: Python, TypeScript, C++, Java, SQL. Learning: PyTorch, scikit-learn, XGBoost/LightGBM/CatBoost, OpenCV, NumPy/Pandas. Serving: FastAPI, Node/Express, PostgreSQL, MongoDB, Docker. Output: React apps, ML microservices, competition-grade models.


⟢ experiments/ — click a run to open the repo

Chest X-ray Classification: 20 classes, DenseNet-201, focal loss, per-class thresholds. Test score −5.22 → −5.04. Mechanical Parts Segmentation: U-Net++, trained from scratch, TTA ensemble. Dice 0.9796. Plant Leaf Super-Resolution: 4× SRGAN, 32 → 128 px. MAE 16.77. Heavy Equipment Pricing: stacked XGBoost, LightGBM, CatBoost on 138K rows. RMSLE 0.1928.


⟢ deploy/ — where the models meet users

AI Transaction Analytics Platform — a full ML product, not a notebook:

flowchart LR
    U([👤 user]) --> R[React + TypeScript<br/>dashboard]
    R -- JWT --> N[Node · Express API]
    N <--> M[(MongoDB)]
    N -- REST --> F{{FastAPI ML service}}
    F --> A[Isolation Forest<br/>anomaly detection]
    F --> S[spend<br/>forecasting]
    F --> T[TF-IDF<br/>auto-categorisation]
    classDef ml fill:#0d1117,stroke:#3fb950,color:#e6edf3
    classDef web fill:#0d1117,stroke:#58a6ff,color:#e6edf3
    class F,A,S,T ml
    class R,N,M web
Loading

AI Mental Health Support Platform — journaling, self-assessments and an OpenAI-powered chat on React · Express · MongoDB with role-based access control.


⟢ git log --oneline --graph achievements

+ * 3a91f2e  (HEAD → main)  🥉 2nd Runner-Up · Adani Innovation Mindstorm 2026
+ |                         built the Isolation Forest + LSTM pipeline for ICS anomaly detection
+ * 7c04d1b  🌱 Open Source Contributor · GirlScript Summer of Code 2026
+ |                         issues, features & PRs through reviewed GitHub workflows
+ * e52b8a0  🏁 Top 40 Finalist · Odoo × MSU FootPrints'25 Hackathon
+ * 1f6c3d9  🤝 Junior Training & Placement Coordinator · FTE, MSU Baroda
+ |                         bridging recruitment drives between students and recruiters
  * 0000000  init: hello, world

⟢ tensorboard --logdir=github  (click to expand stats)

GitHub stats GitHub streak


early_stopping(patience=∞) · open an issue, send a PR, or just say hi — the gradient's always open.

Pinned Loading

  1. chest-xray-disease-classification chest-xray-disease-classification Public

    Chest X-ray classification across 20 thoracic pathologies: studying normalisation, architecture choice, focal loss and metric-aware thresholds under extreme class imbalance.

    Jupyter Notebook

  2. mechanical-parts-segmentation mechanical-parts-segmentation Public

    U-Net++ trained from scratch for 7-class segmentation of mechanical parts, with encoder ensembling and test-time augmentation

    Jupyter Notebook

  3. plant-leaf-super-resolution plant-leaf-super-resolution Public

    SRGAN-style 4× super-resolution of plant-leaf images (32→128 px) with Charbonnier, gradient and perceptual losses and delayed adversarial training

    Jupyter Notebook

  4. heavy_equipment_price_prediction heavy_equipment_price_prediction Public

    Stacked XGBoost, LightGBM and CatBoost ensemble for pricing used heavy equipment from sparse, high-cardinality tabular data, with leak-free target encoding.

    Jupyter Notebook