01 — About
Builder of intelligent + secure systems
I'm a Computer Engineering student at Assumption University with a focus on AI engineering and cybersecurity. I like building intelligent systems — and then figuring out how to break them, before someone else does.
My flagship projects bridge AI and security: NYXUS, a full-stack RAG chatbot with production-grade defenses (prompt-injection scoring, PII redaction, OWASP Top 10 code scanner, audit log), and HORUS, an AI-powered threat intelligence dashboard that pulls live CVE data from NVD and summarizes it in plain Thai/English with Gemini.
Beyond those, I work on classical ML, low-level systems with C/C++ and SYCL, and modern web development with Next.js. I'm currently looking for internships and junior roles in AI Engineering, Security Engineering, or full-stack development.
02 — Featured
NYXUS — Full-stack RAG chatbot with security-first design
An end-to-end Retrieval-Augmented Generation chatbot built to demonstrate production-grade engineering across full-stack, AI, and applied security. Users can chat with a Gemini-powered assistant grounded in PDF documents they upload, scan code for OWASP Top 10 vulnerabilities, and review every privileged action through an audit log.
What it does
- PII redaction across 10+ pattern classes including Luhn-validated credit cards, before any user input reaches the LLM
- Weighted prompt-injection scoring across 7 signal classes (instruction override, role hijack, system-prompt leak, jailbreaks, delimiter smuggling)
- OWASP Top 10 code scanner using structured-output Gemini prompts with severity-rated findings and fix recommendations
- Append-only audit log of every privileged action: auth, conversation CRUD, document upload, chat send, security scan
- Hardened HTTP headers, slowapi rate limiting, CORS allowlists, retry/backoff for transient LLM errors
- GitHub Actions CI running pytest, tsc, and prisma validate on every push
Tech
03 — Featured
HORUS — AI-powered threat intelligence dashboard
A real-time CVE intelligence dashboard for security analysts. HORUS ingests live vulnerabilities from the NVD REST API, summarizes them in plain Thai/English with Google Gemini, and answers follow-up questions through a streaming bilingual chat assistant — with one-click PDF export for offline triage.
What it does
- Live CVE ingestion from the NVD REST API v2.0 with CVSS v3.1 / v3.0 / v2 score normalization and severity-band mapping
- Structured Thai-language summaries (affected systems, attacker capabilities, remediation urgency) generated by Gemini 1.5 Flash
- Streaming bilingual Thai/English security assistant with active-CVE context injected into the system prompt
- Filterable analyst UI — keyword, severity, date range, CVSS score — with paginated CVE table and Recharts severity distribution
- One-click PDF export of filtered CVE reports via jsPDF + jspdf-autotable for offline triage and stakeholder sharing
- Rate-limit-aware NVD fetching with optional API-key elevation (5 → 50 requests / 30 seconds)
Tech
03 — Projects
Other things I've built
GPU parallel computing, and classical ML algorithms from scratch. Detailed write-ups and code on GitHub.
SYCL Parallel Computing
60× GPU speedup on NIST SP 800-22 randomness tests — single codebase, any hardware
Reimplements the NIST SP 800-22 Monobit and Block Frequency tests as SYCL parallel kernels. A single portable codebase runs on Intel CPU, Intel GPU, NVIDIA GPU, and AMD GPU via SYCL/DPC++ — no CUDA lock-in. Achieves 60× peak GPU speedup over the NIST STS serial baseline at 1 GB input with 32 Gbit/s sustained throughput; P-values match the reference within 1×10⁻⁵.
ML Coursework
CNN, PCA, clustering, NLP, and rule mining — classical algorithms from scratch
Six implementations: 120-class dog breed classification via CNN transfer learning, eigenface recognition with manual PCA (NumPy/OpenCV), binary sentiment analysis (Logistic Regression / Naive Bayes, TF-IDF), Apriori market basket mining, K-Means clustering with elbow + silhouette validation, and ID3 decision tree on Titanic survival.
Vaultify
Fintech crypto landing page — frontend only
A crypto custody platform landing page with animated price charts, live market tickers, and glassmorphism UI.
04 — Skills
Languages, tools, and topics
AI / ML
Frontend
Backend
Security
Systems
Tools
05 — Contact
Let's build something.
I'm looking for internships and junior roles in AI, security, and full-stack engineering. If you have a role that fits — or just want to chat about RAG, prompt injection, or parallel computing — reach out.
kansinee.dev@gmail.comBachelor of Engineering · Computer Engineering · Assumption University (ABAC) · Bangkok, Thailand · Expected Graduation: October 2027