Portfolio

Production work, not portfolio fluff.

Every project here shipped. Measured. Handed over. No vapourware, no slides.

Client Work

Government & Public SectorGenerative AI / RAG
94 publications · 32 claims · <3s answers

Connected Knowledge System for a Large Research Organisation

A large research organisation published dozens of policy reports per year — each containing findings, arguments, and recommendations — but they accumulated as PDFs that went unread after launch week. Teams had no way to query institutional knowledge across publications.

94
Publications indexed
<3s
Search time
32+
Claims extracted
Nordic EU
Region
Public SectorKnowledge GraphRAGDocument IntelligenceFinland
Food & Beverage WholesaleFull-Stack SaaS + Integrations
143+ SKUs · Zero manual invoices

B2B Wholesale E-commerce Platform with Automated Accounting

A food and beverage distributor processed wholesale orders manually through phone and email. Tiered pricing (retail, wholesale, distributor), minimum order quantities, case-pack logic, and net payment terms were managed in spreadsheets. Every order required manual invoice creation in QuickBooks, and inventory overselling happened during concurrent checkouts.

143+
Products
55
DB migrations
382
Tests passing
5 types
Background jobs
Next.jsSupabaseStripeQuickBooksB2B E-commerceTrigger.dev

Our Own Systems

FintechFull-stack AI + SaaS
Live · 4 modules

Trading Companion, Discipline System for Retail Traders

Retail and semi-professional traders lacked an all-in-one discipline system for position sizing, risk management, trade planning, and journaling. Existing tools focused on signal generation but ignored the execution and psychology side of trading.

4
Active modules
12 weeks
Time to market
Stripe live
Billing
0
Q1 incidents
FintechSaaSReactSupabaseStripePWA
Enterprise AIAI Agents / Platform Engineering
5 director roles · Zero unchecked sends

DataDiwan OS: Running Our Own Company on a Virtual Team

We ran DataDiwan's own sales, delivery, finance, and marketing operations manually across email, spreadsheets, and disconnected tools. Leads waited days for a first reply. Content creation ate hours per topic. There was no single view of what was happening across the business.

5
Director roles
3
Languages supported
<1 day
Lead response time
0
Unchecked external sends
NestJSNext.jsPythonGeminiAI Agentspgvector

Lab & R&D

HealthcareMachine Learning
28% lower RMSE vs baseline

Epidemic Demand Forecasting for Regional Healthcare

A regional healthcare system in a developing market lacked reliable case forecasting to plan ICU capacity, oxygen supply, and public-health interventions. Off-the-shelf global models did not generalise to local epidemiological data.

4 weeks
Forecast horizon
+28%
Accuracy gain
40+
Dashboard users
15+
Team size
HealthcareXGBoostTime SeriesPublic Health
Computer VisionDeep Learning
4x smaller · 89% accuracy

Emotion Recognition with Mobile-Optimised Deep Learning

State-of-the-art emotion recognition models were too large and power-hungry for on-device mobile deployment. Running inference in the cloud introduced latency, privacy concerns, and dependency on connectivity.

4x smaller
Model size
89%
Accuracy (RAF-DB)
On-device
Inference
Full
Privacy
Computer VisionPyTorchEfficientNetMobile AIQuantisation
Climate TechAI Agents / MLOps
70% faster verification

Multi-Agent Carbon Fraud Detection System

Voluntary carbon markets cannot scale if every project claim and emission sheet requires manual review. Policy-aligned verification checks need to run in parallel across multiple data sources without overwhelming human analysts.

+70%
Verification speed
3+
Data sources
Multi-agent AI
Architecture
Google Cloud
Cloud
Climate TechAI AgentsFastAPIGoogle CloudMLOpsVertex AI
E-commerceBusiness Intelligence
15+ KPIs · Real-time

E-commerce Analytics Platform with CLV & Churn Prediction

E-commerce and growth teams managed multiple KPIs (revenue, CLV, churn, product mix) without a unified interactive platform for cohort filtering and time-range analysis. Spreadsheets could not handle real-time dashboard requirements.

15+
Visualizations
CLV, Churn, Revenue
KPIs tracked
Real-time
Update frequency
All of them
Export rituals replaced
E-commerceBIPlotlyStreamlitCLVChurn Analysis
Enterprise AIGenerative AI / RAG
Cited · Open-source

Enterprise RAG Knowledge Assistant with FAISS

Users cannot safely paste long PDFs into generic LLMs. Enterprise knowledge bases require grounded answers with source citations, and knowledge must update when documents change without retraining entire models.

100%
Cited answers
Gemma-2B
Model
None
Lock-in
HF Spaces
Deployment
Enterprise AIRAGFAISSHugging FaceLLMDocument Intelligence
TransportationMachine Learning
1.4M trips · XGBoost

NYC Taxi Trip Duration Forecasting at City Scale

NYC trip duration is driven by time-of-day, origin-destination patterns, and city-scale congestion. Simple time averages mis-price ETAs in operational routing systems, leading to customer dissatisfaction and driver inefficiency.

1.4M trips
Training data
Temporal, Geo, Weather
Patterns
XGBoost
Model
Streamlit
Deployment
TransportationXGBoostUrban AnalyticsGeospatialTime Series