🧰
Core Languages & Databases
Production-ready code and structured data pipelines.
Python
SQL
PostgreSQL
SQLite
Data • ML • Storytelling
I am Nayan, a data scientist dedicated to transforming complex datasets into actionable insights. I specialize in machine learning, data analytics, and storytelling, building end-to-end solutions—from robust data pipelines and predictive models to interactive dashboards—that help organizations make smarter, data-driven decisions.
Currently
Pune Maharashtra, India
Focus
Production ML Systems
30+
Self Initiated Projects
GitHub
Projects
Connecting the dots between data, engineering, and real-world impact.
Aspiring Data Scientist
Passionate about building intelligent systems that are scalable, interpretable, and impactful.
From NLP to deep learning, I don't just train models—I design pipelines, ensure data quality, and deploy solutions with clean engineering. I bridge the gap between theoretical algorithms and deployed applications.
Full-cycle ML: From raw data collection to Real deployment.
I blend analytical rigor with engineering best practices. I don't just run experiments; I write clean, modular code that is ready for production.
Open to Work
Production-ready code and structured data pipelines.
Exploratory analysis, statistics, and storytelling.
Frameworks to build fast, reliable models.
Classical algorithms for tabular problems.
Neural networks, NLP, embeddings, and clustering
Metrics, serialization, and production patterns.
Connects to data sources and validates input integrity.
Removes noise and prepares structured data for modeling.
Trains the model using optimized feature pipelines.
Measures performance using real validation metrics.
Tracks system health and model stability in production.
VectorCine AI is a High-Fidelity movie recommendation system for vector-based similarity using interpretable modeling.
VERITTA-AI is a High-Fidelity review classification system for noisy real-world text using interpretable NLP.
Mode: Strict
RiskFlow v2.0 is a High-Fidelity churn risk intelligence system for structured banking-style customer data.
Neurovia is a High-Fidelity brain tumor detection system for MRI scans using interpretable, deep learning system.
Due to free-tier infrastructure constraints and inference runtime requirements, a stable public deployment is not currently available.
The system was successfully tested in local and controlled environments. Complete implementation details and design rationale are documented.
Neurovia was evaluated for public deployment under free-tier infrastructure constraints. While the system initializes correctly, the combined requirements of image preprocessing, TensorFlow Lite inference, and server execution introduce stability risks in constrained environments.
Rather than exposing an unreliable diagnostic interface, public deployment has been intentionally deferred. The full system including model execution, inference flow, confidence signaling, logging, and frontend orchestration operates correctly in local and properly provisioned environments (Docker / VPS / controlled cloud instances).
This decision reflects a system-first approach where correctness, transparency, and responsible representation take precedence over forced availability.
INFERSIS-AI is a High-Fidelity news classification system for real-world content using interpretable NLP.
INBOXIS AI is a High-Fidelity Email and SMS classification system for real-world messages using interpretable NLP.
Exploring pricing patterns through business-driven EDA. Uncovering critical factors like target leakage and neighbourhood data quality for robust ML pipelines.
Analyzing donation behaviors to highlight city-level activity and consistency. Emphasizes data quality and documents modeling failures for responsible data science.
Most portfolios are templates. This is a custom-engineered system focusing on performance, scalability, and context-aware interactions.This isn't overengineering; it's craftsmanship.
// System Modules (Active)
State-driven attention control.
Technical Implementation
Context-aware recruiter assistant.
Backend Logic
Scroll inertia & storytelling logic.
Performance Metrics
// Interaction Philosophy
I didn't take the easy way out with a template. Every interaction is built to prove that function implies form.
135 Commits
Interaction Feel
Prioritizing human perception over metrics.
Architecture
Pure Vanilla JS. No framework bloat.
Component Reusability
Modular sections driven by config, not hard-coded spaghetti.
No Heavy Frameworks
Vanilla JS architecture proves understanding of the DOM and core performance principles.
Scalability
Built to expand effortlessly as my project library grows.
Background Transparency
Former MERN Stack Developer
Gave me the discipline to build robust, interactive interfaces.
Data Scientist (Current)
Allows me to build logic that actually parses complex data, not just displays it.
// Engineering Retrospective
React is powerful, but Virtual DOM reconciliation creates overhead for frame-perfect animations.
The Trade-off:
Chose Vanilla JS for direct render cycle control, prioritizing raw performance over dev speed. This helped me to stay close to how the browser actually renders UI.
Managing 11k+ lines of Vanilla JS without a framework led to severe layout shifts and lag.
The Fix:
Refactored game loop to an isolated layer, achieving 60fps stability alongside AI processing. It pushed me to rethink structure and performance in a real-world way.
I pushed this portfolio beyond a static site, building interactive systems, client-side intelligence, and layered motion.
Learned:
Performance is about timing, prioritization, and architecture. I shipped to learn from constraints and iterate in real conditions fast, without sacrificing clarity.
Feature Spotlight
I noticed modern browsers providing smart page summaries. Instead of waiting for API access, I decided to build my own version to elevate the user experience—making it adapting naturally to where you navigate.
I didn't want a generic chatbot. I built a system that feels "aware" of the portfolio's context.
When something inspires me,
I don’t wait for access.
I build my own version and make it better.
Selected Udemy certificates
Issued: July 15 2025
Short: Learned data preprocessing, model building, evaluation, and deployment using Python, Scikit-learn, TensorFlow, and advanced NLP techniques.
Issued: 31 Jan 2025
Short: Gained strong proficiency in writing complex queries, performing joins, subqueries, and advanced data filtering techniques.
Issued: 09 Feb 2025
Short: Built robust ML models using supervised and unsupervised learning techniques with Python and applied them in practical projects.
You might wonder why this portfolio feels so modern and polished — it reflects my experience as a former full-stack developer and my transition into Data Science, combining engineering discipline with machine learning expertise.
JARVIS OS Project
Full-Stack Architecture
MERN Stack Developer
ML & NLP Focus
Responses are tailored to the context of this portfolio.
Let's Connect
Codebase
Click to Launch
Warm Environment
Typical Activity
Analyzing time bucket...