Proof,not portfolio.
A portfolio shows the interface. It doesn't show a row lock contending, traffic doubling overnight, or a region going dark. We show the eight systems built to survive those conditions.
Scroll
For established businesses whose growth has outgrown their foundations.
Manifesto
Frame 01 · Manifesto
0
public benchmark(running in real Workers runtime, not mocks)
You already know what it's like to ship something real. Most founders we work with have built something significant, but their systems don't scale to it yet.
That gap costs more than downtime. It costs the certainty that your architecture will finally survive a sudden viral moment or the volume you never saw coming.
Sohel Ahmed Joni·Founder, LightCavalry
Frame 02 · Verification Harness
Three gates betweenevery commit and production.
Types catch broken contracts. Real-runtime tests catch failure modes mocks can't. Deployment validation catches configuration errors before production. You don't receive a promise that it works: you receive the evidence.
Code that hasn't passed all three isn't done.
CI · 3-gate verification/Zero mocks
~/lightcavalry/ci · 3-gate-verification
Passing
01.
$ wrangler types…Pass0.8s
Regenerates configuration without errors. Binding mismatches die here.
02.
$ vitest run…Pass1.4s
Executes 973 specs inside the real Workers runtime (Miniflare, D1, KV, R2, Durable Objects, Queues). Zero mocks.
03.
$ wrangler deploy --dry-run…Pass0.6s
Validates deployment bindings before anything ships.
0/ 973 specs passing
Runtime: Real edge primitives (zero mocks)
01 / 08
Quorum when the network splits
Raft leader election and heartbeats that keep a cluster coordinated when a node or network link fails—not just when everything is healthy.
// raft · leader election · 150ms heartbeat
02 / 08
Conflict-free state
CRDTs merge concurrent edits by rule rather than a central lock, so replicas can converge after independent changes.
// crdt · merkle merge · no central lock
03 / 08
Atomic commit across partitions
A coordinator drives prepare → commit across partitions, so the transaction is all-or-nothing.
// 2pc · coordinator · all-or-nothing
04 / 08
Getting late events right
Watermarks place late and out-of-order events into the correct time window without forcing the whole stream to be replayed.
// event time · watermark · late / out-of-order
05 / 08
A filter before the database
A Bloom filter checks likely membership before storage is touched, keeping unnecessary reads out of the database.
// bloom filter · k=4 m=64
06 / 08
Change capture, not polling
CDC streams database mutations downstream instead of repeatedly polling for changes, with idempotent handling for at-least-once delivery.
// change data capture · idempotent sink · no polling
07 / 08
Measuring the tail, not the average
HdrHistogram captures the long tail in microseconds, where p99 and p99.9 latency reveal behavior averages conceal.
// hdrhistogram · p50 / p95 / p99 · µs buckets
08 / 08
A graph, not a pipeline
An incremental dependency graph recomputes only the affected downstream nodes, with cycle protection built into execution.
// dag · topological order · cycle guard
Eight systems built around the parts that couldn't afford to fail.
Read the code on GitHubgithub.com/lightcavalry/ddia-on-cloudflare
01 / 08Panel 1 of 8
Scroll →
Frame 04
WE BUILD WHAT SURVIVES
The part of your stack that keeps you up at night is usually the part nobody planned for. We find it, build it, and prove it before your users do.
Frame 05 · Production Blueprints
The systems we engineer.
Blueprint / 01/03
DDIA on Edge Infrastructure
A complete architectural mapping of Designing Data-Intensive Applications to modern edge runtimes. Every chapter, the hard parts included, engineered and verified.
0
verified test specs in a real Workers runtimeConsensus Topology14 DDIA Chapters · 973 Live Runtime Specs
Blueprint / 02/03
Edge-Native Commerce Core
A production-grade blueprint for high-concurrency multi-tenant commerce, from catalog streaming to atomic multi-region checkout.
0
production architecture areas auditedReal-time recommendationsWorkers AI + Vectorize + Queues + KV
Blueprint / 03/03
The Cloudflare Gap List
Ten places where managed cloud providers don't cover the hard part. Each one built, tested against real runtimes, and documented.
0
production gap primitives, each verifiedTHE HARD PARTS CLOUDFLARE DOESN'T GIVE YOU
Consensus, CRDTs, 2PC, CDC, windowing, columnar, graph, sketches, formal methods, mesh. If it's missing, we build it.
01Consensus02CRDTs032PC04CDC05Windowing06Columnar07Graph08Sketches09Formal methods10Mesh
Frame 06 — Capability Map
Six disciplines.Every one mapped to a failure mode.
We don't sell generic development hours. We map engineering disciplines directly to the exact failure modes threatening your business.
- 01
Distributed Systems
resolves “Our database locks up whenever traffic spikes.”core01Distributed Systems
coreresolves “Our database locks up whenever traffic spikes.” - 02
High-Throughput Ingestion
resolves “Events and webhooks are dropping under peak load.”↗02High-Throughput Ingestion
↗resolves “Events and webhooks are dropping under peak load.” - 03
Edge Architecture
resolves “Global latency is killing international conversion.”↗03Edge Architecture
↗resolves “Global latency is killing international conversion.” - 04
State Synchronization
resolves “Users see conflicting state across their devices.”↗04State Synchronization
↗resolves “Users see conflicting state across their devices.” - 05
Security Engineering
resolves “We're not sure we'd survive a rigorous compliance audit.”↗05Security Engineering
↗resolves “We're not sure we'd survive a rigorous compliance audit.” - 06
Architecture Cost Analysis
resolves “Our cloud bill is growing faster than our revenue.”↗06Architecture Cost Analysis
↗resolves “Our cloud bill is growing faster than our revenue.”
Frame 07 · Method & Diagnostics
Four phases.Sixteen failure modes.
Some of the most expensive cloud problems are decided before the first server is provisioned. We diagnose the constraint upstream before writing code, and eliminate the landmines that explode cloud bills at 10x.
01
We find the bottleneck
A short diagnostic that names the actual constraint on your business (data, infrastructure, workflow, payment, auth, reporting, integration, or architecture). You get a one-page brief that names the bottleneck and the system it implies.
02
We engineer the system
With the bottleneck identified, we design the system that removes it. Data model, service boundaries, consistency, failure modes, scalability ceiling, and the cost traps you'd otherwise hit later.
03
We verify and ship
Build the system. Verify continuously against three gates: type checks pass; tests pass inside the real runtime, not mocks; deployment bindings validate. Code that hasn't passed all three isn't done.
three gatesverifying…
- wrangler types: regenerating config
- vitest run: all specs green
- wrangler deploy --dry-run: validating bindings
- Shipping to production...
04
We operate and evolve
Operate the system, or hand it over with a clear operations manual. Identify the next bottleneck. The studio prefers long-term relationships where the engagement evolves as the business evolves.
16 Cost Landmines
- 01Excessive read amplification
- 02Missing/ineffective indexes
- 03Cross-region egress fees
- 04High-cardinality observability
- 05Synchronous RPC cascades
- 06N+1 query patterns
- 07Unbounded result sets
- 08Hot partition keys
- 09Cold-start latency on Workers
- 10Over-provisioned memory limits
- 11Unbatched queue producers
- 12Stale cache invalidation
- 13Schema migration drift
- 14Connection pool exhaustion
- 15Uncompressed large payloads
- 16Idempotency key collisions
ScopeArchitecture audit
Frame 08 · Qualification
Who this is forand who it isn't.
We pick our projects carefully: not to be difficult, but because this is the only area where we can truly deliver exceptional value. If your problem isn't a deep systems challenge, we'll tell you upfront, so you don't waste your budget chasing the wrong solution.
You're a great fit if:
- ✓Your software is business-critical, and your current foundation is starting to crack under the weight of your own growth.
- ✓You're planning for a 5x to 10x jump in traffic, data volume, or concurrent users, and you know your current setup won't survive it.
- ✓You're dealing with real infrastructure headaches: distributed sync failures, unpredictable tail latency, or cloud bills that keep climbing without a clear reason.
- ✓You value hard evidence over slide decks: verified code, real runtime tests, and clear architecture specs matter more to you than reassuring presentations.
- ✓Or, you're building something new and simply want to get the architecture right from day one, creating a highly fault-tolerant system with industry-leading engineering. It'll be more affordable in total cost of ownership than fixing a broken foundation later.
We're probably not the best choice if:
- ✕You need a straightforward CRUD app, an early-stage MVP to test product-market fit, or a marketing-focused site (like WordPress).
- ✕You're mainly looking to expand your team with hourly contractors or general staff augmentation.
- ✕You're currently in the validation phase, where design mocks, feature roadmaps, and promises are more valuable to you than deep, measurable infrastructure verification.
Engagement modelsFixed-fee or retainer · never hourly
01
Architecture Audit
2–4 weeks
A diagnostic constraint report naming the bottleneck + 10x scaling roadmap. Fixed-fee.
02Core Engagement
System Build
3–6 months
Design, build, and verify the production system that removes the bottleneck with 3-gate CI. Fixed-fee.
03
Embedded Partnership
6–12 months
Ongoing architectural stewardship, performance profiling, and system evolution as volume scales. Retainer.
Public system teardown
Send us a system. We'll answer one question for free: what would break if it grew 10x?
No NDA theatre, no sales call. A written teardown of the bottleneck and the landmines we'd scan for, yours to keep.
Questions & answers
Frame 09 · Direct answers
Here's what you need to consider before engaging the studio.
TELL US WHAT BREAKSWHEN VOLUME GROWS 10X.WE'LL TELL YOU WHERE WE'D START.
cavalry@lightcavalry.dev