I'm a third-year Computer Engineering & Data Science student at C.K. Pithawala College (GTU), currently working as a Machine Learning Intern at SAP — I go by "Librarian" online. My focus areas are anomaly detection systems, LLM-driven memory architectures, and turning enterprise data pipelines into something a human reviewer can actually trust.
Outside of internship work, I design and ship independent products end-to-end — browser extensions, forensic deep learning systems, and full-stack ML web apps — with a product engineer's eye for what's usable, not just technically interesting.
Open To:
- Internship / Co-op opportunities in AI/ML, data engineering, or software engineering
- Freelance / contract work — ML pipelines, full-stack builds, developer tooling
Languages
AI / ML
Frontend
Backend
Databases
Cloud & DevOps
Librarian-Code
Local-first CLI coding agent with capsule-based memory, multi-turn conversation, diff previews, custom skills, and multi-provider LLM support. Published on PyPI.
| Stack | Key Features | Repository |
|---|---|---|
| Python, Groq API, FAISS, AST parsing | Capsule memory, codebase-aware context, custom skill system, diff previews | Librarian-Code |
PDF Dark Mode Reader
Manifest V3 browser extension bringing GPU-accelerated dark mode to PDF reading, rebuilt for performance. 250+ pages handled with viewport virtualization.
| Stack | Key Features | Repository |
|---|---|---|
| Manifest V3, GPU CSS filters, Viewport Virtualization | 6 reading themes, async search, glassmorphic UI, 2.1GB → 80MB memory optimization | PDF-DARKMODE |
Footy Prophet
Multi-model football prediction engine combining gradient boosting, neural nets, and classical statistical modeling.
| Stack | Key Features | Repository |
|---|---|---|
| React + Vite, FastAPI, LightGBM, PyTorch, Dixon-Coles | Optuna-tuned (120 trials), 4 seasons EPL/La Liga/UCL data, xG estimates | Footy-Prophet |
Capsule Vault
Context-aware coding assistant — Chrome extension + Python CLI that auto-captures conversations from 8+ AI platforms with codebase-aware context.
| Stack | Key Features | Repository |
|---|---|---|
| Python, Chrome Extension, Groq, FAISS, AST splitting | Multi-platform capture, codebase indexing, local-first privacy | Capsule-Vault |
AI-Forensic
Three-branch deep learning system for detecting manipulated video — separating spatial, frequency, and temporal artifacts.
| Stack | Key Features | Repository |
|---|---|---|
| XceptionNet, SRM filter, BiLSTM/Transformer, EfficientNet | 0.82 AUC cross-dataset (Celeb-DF), 3-branch architecture, custom PyQt5 UI | AI-Forensic |
MentorConnect
Full-stack mentorship platform with dark-mode-first design, Google OAuth, and role-based auth.
| Stack | Key Features | Repository |
|---|---|---|
| Next.js 14, Firebase Auth, Tailwind CSS, Framer Motion | App Router, Google OAuth, dark-mode-first UX, "Terminal Luxury" design system | Mentor-Connect |
XScout
Football scouting analytics dashboard with four analysis modules built on open match-event data.
| Stack | Key Features | Repository |
|---|---|---|
| JavaScript, Python, D3.js, Google Charts, statsbombpy | Role Fit Analysis (7 roles), Similar Player Finder, Euclidean distance matching | XScout |
Study Sync AI
Four-layer study platform pairing a JSF dashboard with a fully local, privacy-respecting RAG pipeline.
| Stack | Key Features | Repository |
|---|---|---|
| JSF 2.2, Jetty 9, Hibernate, MySQL, Qwen 2.5, PageIndex | Local-first RAG, "Viva Mode" oral-exam simulator, auto-generated flashcards | Study-Sync-AI |
April 2026 – Present
Building Data Pristine, an end-to-end data quality and anomaly detection pipeline connecting SAP S/4HANA On-Prem to Cloud Foundry, and designing a capsule-based LLM memory layer for the platform's reasoning components.
- Architected a 6-model anomaly detection ensemble (statistical + ML signal models) feeding a meta-classifier, validated through a human-in-the-loop review layer
- Designed a capsule-based memory system giving LLM components structured, confidence-weighted recall of past anomalies and reviewer feedback, built on SAP's Grounding Management, Orchestration, and Schedules services
- Built a synthetic data generation pipeline producing privacy-preserving enterprise order data, benchmarked against standard utility/privacy tradeoffs
- Worked across SAP BTP, Cloud Foundry, Fiori, and OData to ship a validation layer surfaced through a production-style dashboard
Python LLM Orchestration SAP BTP SAP AI Launchpad SAP HANA Cloud Foundry Fiori
| Recognition | Details |
|---|---|
| Smart India Hackathon | Advanced past the college-level round; led model selection for cattle/buffalo breed classification (90%+ accuracy across 10+ breeds) |
| Code Unnati Innovation Marathon | Semi-Finalist |
current_focus:
learning:
- SAP BTP service architecture & OData
- Capsule-based memory systems for LLMs
building:
- Librarian-Code — local-first CLI coding agent (PyPI)
- Capsule Vault — privacy-first Chrome extension (IndexedDB)
- Capsule-based LLM memory layer on SAP AI Launchpad
exploring:
- Agentic workflows & multi-model routing
- Synthetic, privacy-preserving data generation
open_to:
- Internship / Co-op opportunities
- Freelance & contract engineering work
"Every system I build remembers why it failed — that's how it gets better."
