I work across AI, security, and software engineering — from agents that reason over code to satellite nowcasting, fraud intelligence, and computer vision. I’m co-building OpenSec and bringing research into working systems.
AEGIS connects repository-scale analysis, code retrieval, program-flow reasoning, and agentic triage to proposed patches. TRACEVUL links vulnerabilities to code, fixes, and validation evidence for training and evaluation.
Inside OpenSec: research and intelligence
- AEGIS: 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. The corpus links CVEs to vulnerable regions, fix states, patch evidence, call graphs, root causes, and validation metadata.
- Reasoning model: 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.
- Support: INR 38 lakh Government of India grant for AI-native vulnerability detection and automated remediation.
Project notes and more code
- CloudChase: U-Net cloud coverage prediction using multi-spectral INSAT-3DS imagery, physics-aware losses, SSIM, PSNR, and nowcasting workflows.
- TxnGuard: Secure banking, autonomous fraud detection, and an AML multi-agent framework using vectorized transaction evidence and CrewAI triage.
- Virtual Try-On: Pose-guided try-on with OpenPose, SCW-VTON style warping, skin-tone analysis, and visual retrieval over 7K+ catalog items.
- LeetCode: C++ solutions focused on efficient approaches, pattern recognition, and clean implementation notes.
- Dataset pipeline: More of my data engineering work.
National Winner · ISRO — Bhartiya Antariksh Hackathon 2025.- National Winner · NCIIPC / NTRO — Startup India AI Grand Challenge, Stages 1 and 2.
- Winner · BITS Pilani — Coding Hackathon at APOGEE’25.
- Winner · DTU — VisionX Hackathon at INVICTUS’25.
AI Research Intern · DRDO — Transfer-based black-box adversarial ML for face verification systems.
Research Head · AIMS-DTU — Artificial Intelligence and Machine Learning Society.
Stats refresh daily. Stars earned counts stars on my public, non-fork repositories; repositories starred counts projects I have starred. Contributions follow GitHub’s public calendar. View the dated data snapshot · Refresh status
AI & ML
Languages
Frameworks & data
Cloud & developer tools
AI & ML PyTorch · TensorFlow · Hugging Face · OpenCV · scikit-learn · LangChain · RAG
Systems Python · C++ · FastAPI · Flask · PostgreSQL · MongoDB · Redis · Docker · Linux
Web & cloud TypeScript · JavaScript · React · Next.js · Node.js · AWS · GCP
Security SAST · SCA · DAST · Semgrep · Burp Suite · Ghidra · IDA · CWE/CVE
Full toolkit
Languages: Python, C++, C, Java, JavaScript, TypeScript, SQL, Bash.
AI & ML: PyTorch, TensorFlow, Hugging Face, OpenCV, scikit-learn, LangChain, retrieval systems, embeddings, and LLM pipelines.
Backend & infrastructure: FastAPI, Flask, Node.js, PostgreSQL, MongoDB, Redis, Docker, Linux, AWS, GCP.
Frontend: React, Next.js.
Security & developer tools: SAST, SCA, DAST, Burp Suite, Ghidra, IDA, Semgrep, CWE/CVE mapping, Git, GitHub, Postman.
- LinkedIn: kushalkh
- LeetCode: kushalkhemka
- TryHackMe: kushalkhemka
- GitHub: Kushalkhemka
- OpenSec: opensec.in