I turn ambiguous business problems into pragmatic architectures — clear boundaries, defensible trade-offs, and systems that scale without over-engineering.
- 🏗️ Solution Architect shipping resilient, multi-domain systems — from on-chain analytics to booking engines, marketplaces, logistics and AI automation.
- 🎯 I optimize for the right trade-off, not the trendy one: I distribute only when a single node genuinely can't cope, and I make every architectural decision defensible with an ADR.
- 🌐 10+ domains, 20+ projects, 10+ clients across Germany and Vietnam.
- 📚 Currently deepening data-intensive-systems fundamentals in public — see Labs below.
- ☁️ AWS-certified · PSM I · comfortable from database internals up to org-wide platform strategy.
Architecture & Data
Cloud & DevOps
Languages & Frameworks
Small, runnable repos where I work through core data-intensive-systems trade-offs. Each ships with an ADR (the reasoning) and a measured result (not just a claim).
| Lab | What it demonstrates | Status |
|---|---|---|
| oltp-to-olap-pipeline | Row-store vs column-store, star-schema modelling, and what ETL is really for — with a benchmark you can reproduce | ✅ Live |
| storage-compute-separation | Cloud-native storage/compute split, with a cost model showing when it's cheaper than a coupled DB — and when it isn't | ✅ Live |
| scale-up-before-out | Load-testing one node's throughput ceiling, and the arithmetic for when distributing is actually justified | ✅ Live |
| privacy-by-design-reference | GDPR rights (access, erasure, portability, retention) driven by a declarative PII data map — with verifiable erasure | ✅ Live |
| measure-latency-right | Why the mean lies, why percentiles don't average, and coordinated omission — three runnable latency demos | ✅ Live |
| overload-resilience | Retry storms & metastable failure vs load shedding + circuit breaker; goodput simulated and compared | ✅ Live |
| chaos-failover | Kill a replica mid-traffic: SPOF turns a fault into an outage, failover survives — measured downtime | ✅ Live |
| safe-deploys-canary | Big-bang vs canary + auto-rollback: same broken release, 99.5% less blast radius | ✅ Live |
| choosing-a-database | Pick the model then the engine (SQLite vs Postgres...): measured trade-offs + a decision engine over 16 DBs | ✅ Live |
| lsm-vs-btree | A mini LSM engine built from scratch (Bloom filters, compaction, tombstones) + the measured index trade-off | ✅ Live |
| beyond-the-btree-indexes | Inverted (full-text), spatial (geo), and vector (HNSW) indexes vs brute-force — incl. recall-vs-latency for RAG | ✅ Live |
| replication-consistency | Quantifying "eventual": quorum staleness (w+r>n), async read anomalies, and failover write loss | ✅ Live |
| replication-conflict-resolution | Concurrent writes: LWW data loss vs CRDT convergence, and version vectors for detecting concurrency | ✅ Live |
| partition-key-strategies | The partition key you can't take back: hash vs key-range hot spots, virtual-node skew, and splitting a hot key | ✅ Live |
| shard-rebalancing-and-routing | Scale-out without moving everything: mod N (moves 95%) vs fixed shards vs consistent hashing, plus local vs global secondary indexes | ✅ Live |
| high-demand-ticketing | Selling tickets under a stampede (Laravel): naive oversells 400% vs pessimistic/atomic decrements, TTL holds, idempotent payment, waiting room + AWS on-demand/serverless designs | ✅ Live |
| ecommerce-flash-sale | Why a flash sale oversubscribes the cart & decrements at checkout: funnel vs reserve-at-cart, payment-TTL stock reclaim, per-user anti-scalper cap | ✅ Live |
| multi-resource-booking | Booking across resources/services: naive partial booking vs saga (0 orphans), saga-vs-2PC blocking, hotel date-range atomicity, deliberate airline overbooking | ✅ Live |
📖 Companion write-ups live on longpham.tech/blog.
Designing systems that are simple on purpose. → longpham.tech