Projects
Systems that had to survive contact with reality.
Each initiative paired narrative clarity with operating rigor — from safety and spatial intelligence to customer systems and founder-scale execution.
200+ daily active users
Enterprise supplier compliance & deforestation monitoring
Owned backend architecture and product delivery for a confidential GIS SaaS used by global FMCGs and large agri‑supply partners — 24M+ ha on the map, 200+ daily active users, and roughly 60% faster reporting cycles vs. manual bi‑weekly runs.
DAU
200+
Reporting cycle
~60% faster
Cloud cost
~20% lower
Monitored scope
24M+ ha
Outcomes
200+ daily actives, ~60% faster reporting and data handoffs, ~20% lower recurring cloud spend after architecture hardening — one trusted surface for compliance, not competing spreadsheets.
View project
58,000+ neighborhoods · sub‑second ranking
Neighborhood intelligence & preference‑driven home search
Led product and engineering for a PropTech SaaS that ranks 58,000+ UK neighborhoods in sub‑second time against weighted lifestyle signals — 1,000+ signups in month one, 900,000+ sales records in the scoring layer, and 80% of new users completing a search in their first session.
Launch window
1,000+ users
Index size
58,000+
Sales records
900,000+
Search latency
<1s
Outcomes
Scaled acquisition (1,000+ users in 30 days), hardened search to sub‑second responses at full index size, held 99.5% uptime post‑launch, and shipped pass‑based monetization with automated reporting — a credible story for PM and tech‑lead interviews.
View project
10,000+ users · 100+ exports/day
National geospatial data marketplace & self‑serve exports
Shipped a country‑scale SaaS where planners and GIS teams buy clipped, standardized layers from one catalog — 10,000+ registered users, 100+ automated export jobs per day, and roughly 70% faster data prep vs. manual hunting across government portals.
Registered users
10,000+
Daily exports
100+
Prep time
~70% faster
Repeat usage
90%+
Outcomes
High repeat engagement (90%+ among active buyers), thousands of secure checkouts through a regional payment stack, and a durable pipeline from map AOI → priced clip → SHP/DXF delivery with email receipts and re‑downloadable history.
View project
~90% smaller models · ~3.5× faster inference
Edge environmental & fire intelligence — ML optimization and Earth‑observation validation
Partnered with a deep‑tech hardware startup to harden a proprietary badge‑class edge AI stack for real‑time environmental and fire‑risk signals: structured pruning and INT8 / INT4 quantization for on‑chip inference, sensor stabilization under temperature and humidity drift, manufacturing‑neutral baselines, gradient‑based input reduction, and VIIRS / GOES‑16 fusion with NASA FIRMS for high‑confidence event validation — roughly 90% smaller models with about 3.5× inference speedup while tightening operational trust for field teams and exec reviewers.
Model compression
~90%
Inference speedup
~3.5×
Quantization
INT8 / INT4
EO products
VIIRS + GOES‑16
Outcomes
~90% model compression with ~3.5× faster on‑device inference, materially lower memory and I/O after quantization, measurably calmer sensor traces across devices and weather swings, and a defensible satellite cross‑check story for PMs and tech leads hiring into safety, climate, or edge ML.
View project