New systems and services
Embedded senior engineering for a defined window: design docs, code, non-functional tests, handoff.
WEEKS TO MONTHS · IN YOUR REPOSenior software engineering · Production systems
Senior engineering for systems where production behaviour matters.
SteerScale designs, builds and operates backend platforms, distributed systems, cloud infrastructure and AI systems — software that has to hold under real load, real failure and real cost.
Hands-on, senior-engineer-led, measured rather than guessed.
PRODUCTION SYSTEM · ORDERS PLATFORM HEALTHY · 3 AZ
EDGE
api-gatewayauth · rate limit 2k rps · idempotency
SERVICES
order-svckotlin · spring · ×12
pricing-svcjava · webflux · ×6
ai-assistllm router · evals · ×3
EVENTS
kafkaorders.v2 · pricing.changed · 24 partitions · lag 0
DATA
postgresqlorders · primary + 2 replicas
cassandraevents · RF 3 · 80k w/s
redispricing cache · hit 97%
PLATFORM
OBSERVABILITY
P95 LATENCY
41ms
AVAILABILITY 30D
99.95%
WRITE THROUGHPUT
80K/s
ERROR BUDGET
62% left
AI COST / WORKFLOW
€0.09
traces · metrics · logs
otel → prometheus · grafana
otel → prometheus · grafana
15+ YEARSProduction engineering
16MConcurrent users at peak
80K TPSSustained write throughput
EXPERIENCESky · Tesco · ClearScore · Lufthansa Systems · Gamesys
01 / Engineering capabilities
Four disciplines. One senior engineer across all of them.
SVC-01
Backend & Distributed Systems
APIs, event-driven architecture and high-throughput services that behave predictably under real load.
Java · Kotlin · Scala · Kafka · Cassandra · PostgreSQL Backend & Distributed Systems → SVC-02Cloud, Platform & Reliability
Kubernetes and cloud platforms where capacity, latency, observability and resilience are measured targets.
AWS · Azure · Kubernetes · Terraform · OpenTelemetry Cloud, Platform & Reliability → SVC-03AI Systems Engineering
LLM and agent systems run as production software — evaluation, observability, reliability and economics designed in.
Orchestration · RAG · routing · evals · tracing AI Systems Engineering → SVC-04Architecture & Technical Leadership
Architecture reviews, decision records, delivery leadership and hands-on mentoring for teams shipping hard systems.
ADRs · reviews · roadmaps · tech lead cover Architecture & Technical Leadership →02 / How we engage
Build, improve or diagnose. Always with a written scope.
Every engagement starts with a short technical conversation and a one-page proposal: goals, deliverables, cadence, fixed price or time-and-materials, exit criteria. The code lands in your repository.
Existing production systems
Performance, reliability, scalability, cost and operability work on software that is already live.
MEASURED BEFORE AND AFTERFixed-scope assessments
Evidence-driven reviews of one system with quantified findings — such as the AI Systems Audit.
FIXED PRICE · 2–3 WEEKS03 / Featured productized engagement
AI Systems Audit
For teams whose LLM or agent workflow has reached production: a fixed-scope, evidence-driven analysis of one real workflow across economics, reliability, observability and evaluation.
Quantified findings ranked by impact, controlled comparisons where practical, and a prioritized remediation plan your engineers can act on without us.
About the audit Fixed price · Fixed scope · 2–3 weeks
ILLUSTRATIVE OUTPUT — NOT A CLIENT RESULT FINDING 3 OF 11 · WORKFLOW #A83F
High-cost model used for deterministic classification
EVIDENCE
38% of inference calls perform a bounded classification task (12 labels, deterministic output schema) on the same model used for open-ended reasoning. Output agreement with a smaller candidate model was measured on 4,200 replayed production requests.
| METRIC | CURRENT | CANDIDATE |
|---|---|---|
| Quality / agreement | 99.0% | 98.8% |
| Cost / request | €0.018 | €0.003 |
| p50 latency | 1.8s | 0.7s |
POTENTIAL IMPACT
~€4,700/ month
≈ 83% of current classification spend
ENGINEERING EFFORT
LOWCONFIDENCE
HIGHREMEDIATION PRIORITY
Route bounded classification to the candidate model behind a quality gate; keep the current model as fallback below 98.5% agreement.
Subject to production validation
04 / Experience
Built on 15+ years of production engineering.
Streaming platforms serving millions of concurrent viewers. Retail pricing and payment backends. Fintech microservices, airline operations control, real-time gaming platforms. Different domains, the same engineering problems:
load · concurrency · retries · state · latency · failure domains · observability · deployment safety · cost
That experience now also covers AI systems — LLM orchestration, model selection, inference economics and evaluation — as one more workload that has to run reliably inside a distributed system.
16M
concurrent users — live streaming, NFL playoff broadcast
80K
writes / second sustained — bookmarking API on Cassandra
SELECTED PRODUCTION WORK
- Sky / NBCUniversalGlobal video streaming · high-throughput backend
- TescoRetail pricing & payment platforms
- ClearScoreFintech · reactive microservices
- Lufthansa SystemsAviation · operations control
- Gamesys / Ladbrokes CoralHigh-traffic gaming & real-time platforms
Client names shown typographically; full work history on the About page.
The engineer you speak to is the engineer doing the work.
SteerScale is a small, senior-engineer-led practice. The person scoping the work is the person writing the code, reading the traces and presenting the results. One or two engagements at a time, written status every week.
No sales-to-junior handoff.
Barnabás Szőke
Senior Software Architect / Founder
15+ years building and operating backend, distributed and cloud-native systems across streaming, retail, fintech, aviation, gaming and SaaS — Java, Kotlin and Scala services, Kafka, Cassandra, Kubernetes on AWS and Azure.
Current specialization: AI systems in production — orchestration, model selection, inference economics, evaluation and observability — built on that distributed-systems foundation.
FAQ
Questions engineering leaders ask first.
Neither in the usual sense. SteerScale is a senior-engineer-led consultancy: one principal engineer, one or two engagements at a time, embedded in your repository and standups for a defined window with a written scope and exit criteria.
Backend platforms, distributed and event-driven systems, cloud and Kubernetes infrastructure, and AI systems in production — typically where throughput, latency, reliability or cost has become a real constraint.
With a 20–30 minute technical conversation and a one-page proposal: goals, deliverables, cadence, fixed price or time-and-materials, and clear exit criteria. Fixed-scope assessments such as the AI Systems Audit have a fixed price agreed before work starts.
No. AI systems engineering is an important current specialization, but most of the practice is backend, distributed-systems and cloud engineering. AI systems are treated as one more workload that has to run reliably inside those systems.
You do. Code lands on your branch in your CI, decisions are written down as ADRs, and handoff means your engineers are comfortable making the next change without us in the room.
Have something difficult in production?
Bring one real system and the problem behind it — a service that misbehaves under load, a platform that needs to scale, an AI workflow whose cost or reliability nobody can explain. The first conversation is a short technical call.
Discuss your system hello@steerscale.com
20–30 minutes · Technical conversation · No sales presentation