I'm a second-year B.Tech AI & Data Science student at SVKM's NMIMS Mukesh Patel School of Technology Management & Engineering, Navi Mumbai (Class of 2029), currently holding an 8.37 CGPA (First Year).

I'm focused on becoming an AI Engineer — not just building models, but identifying real problems, designing systems around them, and shipping working products. My workflow leans heavily on AI-assisted rapid prototyping: I use tools like agentic coding assistants to go from idea → working system fast, then dig into the parts that matter most (architecture, logic, and decision-making). It's how most of the projects below came together — I can walk through the system design and reasoning behind each one, even where I'm still building up hands-on fluency in every framework used.

Right now, I'm strengthening my core fundamentals — DSA, Python, and backend engineering — so the systems I design are backed by solid implementation skill, not just good ideas.

Following a structured AI/ML roadmap:

• 🐍 Deepening Python fundamentals
• 🧠 Strengthening Data Structures & Algorithms
• 🔢 NumPy & 🐼 Pandas for data handling
• 🤖 Scikit-learn for core ML algorithms
• 📊 Matplotlib for data visualization
• ➡️ Next up: core Machine Learning concepts (regression, classification, model evaluation)

Languages

Libraries (basic proficiency)

Developer Tools

Currently Learning

• Data Structures & Algorithms
• Core Machine Learning concepts

Most of these were built rapidly using AI-assisted prompting (idea → working prototype) during hackathons or self-driven practice. They reflect how I approach problem framing and system design — not polished, production-hardened code. Some parts may be incomplete or rough around the edges.

Built with a 4-person team during HackUp 2026 — placed in the Top 45 out of 200 teams. A prototype for real-time transaction fraud detection, combining behavioral signals with a hybrid ML + rule-based scoring approach and a simple decision engine (Approve / MFA Hold / Block).

• Tech (attempted): FastAPI · Scikit-learn (Random Forest) · React + Vite · SQLite
• Key Learning: Exposure to hybrid scoring system design (ML + rules) and how a real-time decision engine is structured.
• Status: Core backend logic in place; frontend still being reworked and not fully polished.

Won 3rd Place at an Ideathon organized by Tech Cider. A simulation prototype exploring how urban transport policy decisions (train/bus expansion) could be stress-tested hour-by-hour and year-by-year, to surface when and why a system might fail rather than just whether a metric improves.

• Tech: Wolfram Language (Mathematica) — Manipulate, DynamicModule, DensityPlot
• Key Learning: Framing a vague policy question as an explicit, assumption-driven computational model.

An experiment in aggregating company signals (news, tech stack, hiring trends) in parallel and using the Anthropic Claude API to synthesize them into a pre-call sales brief.

• Tech: Python · Streamlit · Anthropic Claude API · SQLite · Jinja2 · WeasyPrint
• Key Learning: First exposure to multi-agent style architecture and structuring LLM prompts for consistent, structured output.
• Note: Prototype-level; not stress-tested end-to-end.

A dashboard concept covering tax regime comparison, a financial health score, FIRE planning, and AI-assisted Form 16 parsing.

• Tech: Streamlit · Plotly · Google Gemini · Groq · FPDF2 · PyMuPDF
• Key Learning: Exposure to combining rule-based financial calculations (tax, SIP, FIRE) with AI-powered document parsing.
• Note: Prototype-level; calculations haven't been independently verified against a CA/financial advisor.

A habit app concept built around execution over tracking — surfacing only today's top-priority habits and allowing recovery from missed ones, with no backend and all data kept on-device.

• Tech: Next.js · TypeScript · Tailwind CSS · Framer Motion · Service Workers (PWA)
• Key Learning: Exposure to history-first state design (deriving streaks/insights from an event log instead of stored flags).

An academic project (Elements of Biology course) that encodes text into DNA base sequences (A/T/C/G) and decodes it back, loosely mimicking DNA's complementary base-pairing.

• Tech: HTML5 · CSS3 · Vanilla JavaScript
• Key Learning: Translating a biological concept into a simple working encode/decode model.

An early-stage, structured practice repository for core data science fundamentals — array operations, broadcasting, and DataFrame manipulation.

• Tech: Python · NumPy · Pandas · Jupyter Notebook
• Status: Just getting started — foundational concepts only for now.

Rather than lean on third-party stat-card services (which kept breaking), this profile relies on GitHub's own native features, which never go down:

• Pinned Repositories — go to your profile → Customize your pins → pick up to 6 (e.g. FraudSense, Universal What-If Engine, DealRadar). These render as GitHub's own repo cards directly on your profile.
• Contribution graph — GitHub already shows this natively at the top of every profile page. No extra widget needed.

• 🥉 3rd Place — Ideathon organized by Tech Cider (Project: Universal What-If Engine)
• 🚀 Top 45 / 200 Teams — HackUp 2026 (Project: FraudSense)
• 🎓 Co-Head, Tech Cider
• 📘 Microsoft AI Skills Fest — Participant

Phase 1 — Foundations (in progress)

• Python (deepening core concepts)
• Data Structures & Algorithms
• NumPy
• Pandas
• Scikit-learn
• Matplotlib

Phase 2 — Up Next

• Core Machine Learning concepts (regression, classification, model evaluation)
• More advanced Python