Oil and Natural Gas Corporation (ONGC)

Summer Intern | Database & Security Group · May 2025 – June 2025

Developed an ML-based Network Intrusion Detection System (NIDS) to address the limitations of static signature-based security. Engineered a data pipeline processing the UNSW-NB15 dataset (2M+ records) and optimized detection models (RandomForest, XGBoost) using SMOTE to handle class imbalance. Analyzed zero-day attack patterns under the guidance of the Database Programming division.

Institution's Innovation Council (IIC) — FRCRCE

IPR Activities Coordinator · Apr 2026 – Present

Coordinating IPR, patent filing, and research commercialization initiatives to foster an innovation ecosystem among engineering students.

TISD-Edu-AI: Local-First RAG Tutor

Privacy-first, on-device AI tutor for Grade 1–10 students built for Apple Silicon M4. Ingested 4,656 pages of NCERT and encyclopedia content into 6,422 semantic chunks across a 2.4M-point vector space (6422 × 384 dimensions). The Phi-3 (3.8B) inference pipeline achieves 41.42 tokens/sec at 1.72s latency with a 9.99GB memory footprint — zero data leaves the device.

gemma-m4-agent: Production Local AI Agent

Production-grade local AI agent implementing Hybrid-Memory Orchestration (HMO) on Apple M4 — 16GB Unified RAM for inference, external APFS drive for long-term vector memory and session state. Runs Gemma 4 8B (8-bit quantized) via MLX with real-time SSE token streaming and DuckDuckGo-grounded web search. Solved LLM hallucinated turn-leakage via dual-gate stop logic: silicon-level token ID injection + software-level rolling string-match guard.

MendikotZero: AlphaZero RL Agent

Engineered an autonomous AI agent for the Indian card game 'Mendikot' using Deep Reinforcement Learning. Implemented the AlphaZero architecture from scratch, utilizing Monte Carlo Tree Search (MCTS) guided by a Dual-Head Neural Network. The agent was trained via self-play without human data, evolving from random moves to advanced strategic planning.

CS50 Cybersecurity: Anatomy of a Critical Failure

CVE-2025-61882 — Oracle Agile PLM (CVSS 9.8)

Technical forensic analysis of a critical Path Traversal vulnerability in Oracle Agile PLM. Deconstructed the input validation failure and proposed defense-in-depth remediation strategies including Path Canonicalization and Least Privilege architecture.

Network Intrusion Detection System (NIDS)

Developed a high-performance NIDS using machine learning to detect novel and zero-day attacks. Executed a full data science workflow on the UNSW-NB15 dataset (2M+ records). The final Random Forest model (with SMOTE) achieved an excellent balance between high detection rates and low false positives.