Projects
Everything here runs
Live demos, not mockups. Each project has a dedicated page covering approach, architecture, tradeoffs, and security decisions.
Invoice Parser
Upload a PDF invoice, get structured JSON: vendor, line items, tax, totals. 3-tier extraction: layout analysis, regex patterns, LLM fallback. Every field schema-validated.
Extraction
Rule-based + LLM
Accuracy
94%
Output
Validated JSON
Security
Sandboxed + rate-limited
Email Classifier
Classify inbound email into support, sales, spam, or internal. Two-stage pipeline: rule-based spam filter then TF-IDF + LinearSVC classifier. No external API calls.
Model
TF-IDF + LinearSVC
F1 score
0.91 weighted
Latency
< 50ms
API calls
None (local model)
Ticket Router
Paste a support message, get it routed: urgency, department, and a suggested first response. Hybrid rule + sentiment + LLM pipeline with PII redaction.
Routing
Urgency + department
Accuracy
89%
PII handling
Redacted before LLM
Escalation
Human-in-the-loop
Sentiment Dashboard
Upload customer reviews or messages, get a sentiment breakdown: positive/negative/neutral percentages, trend chart, and flagged negatives. VADER + optional LLM hybrid.
Throughput
~2,000 msg/sec
Accuracy
91% (hybrid)
Max batch
50,000 messages
Data stored
Never
Demand Forecaster
Interactive forecasting: adjust season, price, promotions and see predicted daily sales shift. Gradient Boosting model with confidence intervals and feature importance.
Model
Gradient Boosting
MAPE
8.2%
Inference
< 20ms
Dependencies
None (local model)
Document Q&A
Upload a PDF, ask questions in plain English, get answers with page-number citations. RAG pipeline: chunk, embed, retrieve, generate. Grounded answers only.
Pipeline
RAG (retrieval-augmented)
Vector DB
ChromaDB
Session expiry
30 min auto-delete
Max document
200 pages
Spreadsheet Reconciler
Upload two CSVs (bank statement vs ledger), get a reconciliation report: matched rows, unmatched rows, discrepancies. Handles 100k+ rows with fuzzy matching.
Matching
Exact + fuzzy
Max rows
100,000 per file
Data stored
Never
CSV injection
Prevented
Market Snapshot
Fetches Cisco's daily high/low and streams a live current price. Runs as a scheduled pipeline with a web view and JSON API.
Schedule
Daily
Live price
Real-time
Output
Web + JSON
Source
yfinance
GlobeScraper
Full-stack platform for English teachers in Southeast Asia. Community, AI blog engine, rental marketplace with a 7-source scraping pipeline.
API routes
55
DB models
30+
Scraper sources
7
Deploy
Vercel + Hetzner
Weather ML App
A machine learning Flask app that predicts weather conditions from sensor inputs. Dockerised with full CI/CD and Kubernetes deployment.
Model
scikit-learn
Deploy
Docker + K8s
CI/CD
GitHub Actions
Tests
smoke + unit + integration
Server Pulse
Live dashboard showing real-time CPU, RAM, disk, and uptime from a Hetzner CX23. Polls a Flask + psutil endpoint every 5 seconds.
Backend
Flask + psutil
Polling
5 seconds
Server
Hetzner CX23
SSL
Let's Encrypt
Uptime Monitor
Background thread pings the portfolio API and site every 60 seconds. Graphs 24 hours of response times with Canvas charts.
Ping interval
60 seconds
History
24 hours
Backend
Flask + threading
Chart
Canvas (zero deps)
Image Style Transfer
Upload any photo and apply a famous art style: Van Gogh, Munch, Picasso, and more. Fast neural style transfer running on the server with OpenCV DNN.
Models
7 styles
Backend
Flask + OpenCV
Inference
CPU only
Max input
10 MB
Security Scanner
Enter any URL and get a full security header audit: HTTPS, HSTS, CSP, clickjacking protection, and more. Live tool with A–F grading, SSRF-safe API, and detailed per-header findings.
Headers checked
9
Grading
A–F
SSRF protection
Yes
Latency
< 2s
Network Health Monitor
Add up to 5 endpoints and monitor them in real time. Polls every 15 seconds, shows status, response time, and SVG sparkline latency charts. SSRF-safe API with batch health checks.
Max endpoints
5
Poll interval
15s
History depth
40 checks
Chart
SVG sparkline
Data Pipeline
End-to-end data pipeline: ingest, validate, transform, and output structured data. Designed for reliability with retry logic and monitoring.
Pattern
ETL pipeline
Validation
Schema + runtime
Monitoring
Structured logging
Retry
Exponential backoff
Want something like this built for your business?
I'll look at your problem, figure out the right approach, and ship working software. No slideshows.
Book a free consultation