Open to a long-term, full-time contract (B2B).
Svetislav Ćirić
Software Engineer. I build the core backend systems of early-stage and scaling startups: from the first commit to production scale, owning the hard technical decisions, and staying to evolve them.
Go and TypeScript. Data-heavy, real-time and correctness-critical systems. Based in Belgrade, Serbia (CET), remote, overlapping US time zones.
What startups hire me for
Build the first version
A contextual ad server, an AI workflow engine, a retail ERP, a social-graph recommender. I ship v1 fast, then make the foundations last.
Fix what is starting to break
Prototypes that need a rewrite, code that drifted after fast iteration, legacy apps nobody understands. I find the real bottleneck and make the smallest change that fixes it.
Connect data across systems
Warehouse imports and exports (BigQuery, Snowflake, Redshift, Databricks, S3, Azure), event pipelines, partner integrations. No lost or double-counted rows.
How I work
- I own outcomes, not tickets. I do my best work close to the founders or CTO, with real agency over a core system.
- I write decisions down. Architecture decision records, design docs and test plans, so the system doesn't depend on my memory.
- I trade on purpose. My biggest wins came from accepting a known cost: memory for latency, freshness for throughput, a manual run for a deadline.
- I use AI to go faster without lowering the bar. It lets me do the work that used to get cut: tests, docs, security review. The architecture is still mine.
- I stay. I owned my longest engagement for four years and stayed on as an advisor afterwards.
Selected stories
One PR, 5–7× faster ad serving
An in-process cache in front of eight Redis-backed caches cut p90 latency from ~150 ms to 20–30 ms, trading memory per pod for an order of magnitude in latency.
Reading the traffic, not the benchmarks
90% of ad traffic came from articles under 14 days old. Splitting search into a hot and a long-tail index cut query time by another order of magnitude.
200 TB through a pipeline built for gigabytes
Sharded a huge cloud-storage import across 192 workers and pushed filters agreed with the customer into DuckDB, finishing an import the existing system could not run.
Let's talk
If you're an early-stage or scaling startup with a hard backend problem and room for someone to own it, email me at[email protected].