chansg
Cyber Security Analyst at Sparta Global, based in London, UK — working across SOC, threat detection, and identity & access management (Entra ID). My foundation is IT engineering: identity-lifecycle automation in PowerShell, Conditional Access, Intune, Zscaler, and Mimecast, on top of a BSc in Computer Science and a completed data bootcamp (May 2026). Longer-term, I’m building toward Detection Engineering.
Security
🔎 AI Security Log Analyser
An anomaly-detection system combining an Isolation Forest model with a rule engine to flag suspicious patterns in authentication telemetry, surfaced through a Django dashboard. A practical exercise in unsupervised ML, feature engineering, and turning model output into something a human can actually triage.
What it shows: unsupervised anomaly detection, feature engineering, and operationalising model output into a usable interface.
Tools: Python · scikit-learn · Django · pandas
Engineering range
📊 BDO Analytics
A real-time market-intelligence dashboard ingesting live Central Market data, surfacing best-seller rankings, price-stability scoring, and anomaly detection across a volatile, high-frequency dataset. Built as an end-to-end analytics product: ingestion → metrics → interactive dashboard with watchlists and CSV export.
What it shows: real-time data ingestion, time-series analysis, anomaly detection, and self-serve dashboarding.
Tools: Python · pandas · Plotly · Streamlit
🎬 Shorts Pipeline
Turning long gameplay recordings into polished short-form video normally takes hours of manual editing; I automated the full stack — CV-based highlight extraction, Whisper speech-to-text with per-speaker diarisation, AI image and video generation, caption burning, and audio mixing — behind a Gradio human-review gate, validated across 295 passing tests.
Tools: Python · ffmpeg · Whisper · Gemini API · ElevenLabs · OpenCV
🤖 Project Aria
A Windows voice assistant reframed as a personal study tool: it records and locally transcribes meetings with Whisper into timestamped, summarisable notes, and answers questions about whatever is on screen — “what’s wrong with line 6?” on a SQL query — by reading the active window. A self-registering capability router (keyword → local semantic embedding → Claude) keeps every action observable and human-approved — side-effecting intents are deterministic keyword-only, so they can’t fire off a loose semantic match — backed by 441 passing tests.
Tools: Python · Claude API · Gemini · faster-whisper · Kokoro ONNX · SQLite
Skills
| Identity & Access | Endpoint, Network & Email | Automation | Data & Analysis |
|---|---|---|---|
| Entra ID, Active Directory | Intune — endpoint management | PowerShell — JML / identity-lifecycle automation | Python — pandas, scikit-learn |
| Conditional Access policies | Zscaler — network & web security | Git, GitHub, PyCharm | SQL — GROUP BY/HAVING, subqueries |
| Joiner–mover–leaver process | Mimecast — email security | Power BI — DAX measures · Excel — INDEX/MATCH |
Where these techniques show up
| Technique | Where it's proven |
|---|---|
| Time-series analysis & anomaly detection | BDO Analytics — live market metrics, price-stability scoring, outlier flagging |
| Unsupervised ML & feature engineering | AI Security Log Analyser — Isolation Forest over authentication telemetry |
| DAX measures, KPI cards, drill-through | Power BI reports — Desktop through to Service publication |
| GROUP BY/HAVING, subqueries, schema design | SQL queries — MySQL Workbench |
| Calculated fields, filters, parameters | Tableau dashboards — interactive layouts on Tableau Public |
| INDEX/MATCH, SUMIFS, Pivot Tables | Excel projects — summary reporting from raw datasets |
Activity
