Built LLM-driven pentesting tooling that automates reconnaissance, test-case generation, exploit validation, evidence collection, and report writing. Chained model calls behind deterministic guardrails and cached context aggressively, reducing assessment timelines from roughly a week to hours at about 10% of standard LLM token cost.

Spearheaded Onchain offensive security assessments after identifying the lack of a standardized pentest path. Integrated threat modeling into repeatable attack-surface discovery and mapped Web2-to-Web3 integration points, maturing coverage from an undefined baseline to 70% across DeFi products and Onchain systems.

Built Java + Jazzer fuzzing pipelines for the AI Cyber Challenge targeting XSS, command injection, and insecure deserialization. Produced reusable fuzz harnesses, seed corpora, and crash-triage automation to let autonomous systems find and reproduce memory- and injection-class bugs without human drivers.

Reverse-engineered ransomware and malware samples with IDA Pro, Ghidra, and x64dbg to extract exploit techniques, persistence mechanisms, and evasion tactics, mapping IOCs and TTPs to MITRE ATT&CK for detection engineering. Built Python tooling and sandbox instrumentation to automate unpacking, string/IOC extraction, and triage across sample batches.

Engineered custom EDR and WAF evasion payloads and obfuscated command-and-control tradecraft to validate detection coverage during adversary simulations. Exposed gaps in endpoint telemetry across 15+ application and infrastructure assessments and drove new detection logic with the blue team.

Innovated a human-solvable CAPTCHA system in Golang to bolster security against offline dictionary attacks. The system featured three unique puzzles: Sudoku, Chess, & Cryptographic Hash Puzzle. Performed comparative analysis with cutting edge solutions leading to our finding of 30% less likelihood of successful dictionary attacks against offline files.

Trained and tuned a deep-learning classifier over chest X-ray imagery, handling class imbalance and augmentation to improve detection of disease markers, and published the notebook and results on Kaggle.

Best Paper Award at IEEE CCWC 2024. I implemented an ML security monitor for maritime ICS infrastructure, achieving 98.5% accuracy in detecting cyber threats on ICS networks and validating its effectiveness through penetration tests against power systems and weapon controls.

Developed proof-of-concept exploit achieving root access by reverse engineering Angband game binary to locate format string vulnerability. Used IDA Pro and GDB to analyze vulnerability and craft input to exploit stack overflow, redirect code execution flow, and open remote shell.

Performed security testing of OpenEMR, identifying vulnerabilities like SQLi, DoS, buffer overflows, and XSS. Used tools including Burp Suite, OWASP ZAP, SQLMAP, and Wireshark to detect issues. Documented proof of exploitation along with remediation recommendations.

Exploited buffer overflow in Duke Nukem game binary to achieve remote root shell access. Reverse-engineered binary using Ghidra to identify vulnerable function and crafted malicious input. Developed proof-of-concept demonstrating arbitrary code execution via shellcode injection and redirecting control flow using buffer overflow technique.

Conducted threat modeling on an open-source web server using SciTool Understand and Microsoft Threat Modeling Tool. Produced an executive summary detailing risks and employed Ghidra and Veles for reverse engineering, bolstering the system's security posture.

Simulated DDoS attack in IoT devices, like flooding on CoAP network using Cooja simulator, leveraged the simulation data to train an ML model for detection of unusual traffic.

Designed and built a tamper-proof, decentralized e-voting system on the Ethereum blockchain through commit-reveal mechanisms. Hardened system security by building Solidity smart contracts with audited OpenZeppelin libraries, optimized consensus algorithms, IPFS integration and MetaMask identity management resulting in enhanced transparency, scalability, and reduced gas costs by 8%.

Investigated anonymized card data, identified feature relationships through visualizations, and engineered ensemble models to detect credit card fraud with 99.9% accuracy, reducing false positives by 20%.

Developed blockchain-based supply chain system using Solidity, OpenZeppelin, Truffle, and Web3, integrating web interfaces for server-chain communication. Enhanced efficiency by 10% and improved security.

Developed a MERN-based web application for integrating news from assorted sources of varying genres by applying several information retrieval techniques, resulting in 30% reduction in similar sentences by applying cosine-based similarity detection.

Developed an Information-Retrieval System for a large database of books with user-friendly interface using Flask. This gives an easy and extended version to library-management systems built for large scale institutions.

Developed and presented a management system for apartments in the area, facilitating easy communication, complaint and announcement forums, and other utilities to provide a smoother management experience for residents of a community.