I build systems where AI agents reason over code, security evidence, and large-scale repositories. My current center of gravity is OpenSec, an AI-native OSS security platform for vulnerability detection, triage, remediation, and secure code intelligence.
|
From vulnerability to visibility. We scan, triage, and remediate vulnerabilities in open-source repositories with AI-powered code intelligence. |
|
| System | Signal |
|---|---|
| AEGIS | Repository-scale detection and remediation using SAST signals, RAG code retrieval, program-flow reasoning, exploitability analysis, multi-agent triage, and automated patch generation. |
| TRACEVUL | 94,759 canonical vulnerability records, 184K training samples, 19+ languages, 553 CWEs, and 6,192 OSS repositories. |
| Reasoning LLM | In-house Qwen 3.6 27B vulnerability reasoning model fine-tuned with RsLoRA SFT for detection-to-remediation workflows. |
| Security research | 28 zero-day vulnerability candidates across 6 OSS repositories, including credited discovery of CVE-2026-42766 in OpenSSL, with accepted fixes in OpenSSL and Vim. |
| Grant | INR 38 lakh Government of India grant for AI-native vulnerability detection and automated remediation. |
|
SAST, SCA, DAST, CWE/CVE mapping, exploitability reasoning, validation, and patch workflows. |
Computer vision, adversarial ML, retrieval systems, embeddings, U-Net nowcasting, and LLM pipelines. |
|
FastAPI, Flask, PostgreSQL, MongoDB, Redis, Docker, Linux, AWS, GCP, and production API surfaces. |
C++ DSA practice, efficient solution patterns, and maintainable implementations in public repositories. |
|
Agentic vulnerability detection, triage, exploitability reasoning, and patch generation for OSS repositories. |
Training and evaluation corpus linking CVEs to vulnerable regions, fix states, patch evidence, call graphs, root causes, and validation metadata. |
|
U-Net based cloud coverage prediction using multi-spectral INSAT-3DS imagery, physics-aware losses, SSIM, PSNR, and nowcasting workflows. |
Secure banking, autonomous fraud detection, and AML multi-agent framework using vectorized transaction evidence and CrewAI triage. |
|
Pose-guided try-on with OpenPose, SCW-VTON style warping, skin-tone analysis, and visual retrieval over 7K+ catalog items. |
C++ solution repository focused on efficient approaches, pattern recognition, and clean implementation notes. |
| National Winner Bhartiya Antariksh Hackathon 2025 by ISRO |
National Winner NCIIPC Startup India AI Grand Challenge Stages 1 and 2 by NTRO, Government of India |
| Winner Coding Hackathon at APOGEE'25 by BITS Pilani |
Winner VisionX Hackathon at INVICTUS'25 by Delhi Technological University |
| Research AI Research Intern at DRDO working on transfer-based black-box adversarial attacks for face verification systems |
Leadership Research Head at AIMS-DTU, Artificial Intelligence and Machine Learning Society |
