Writing
Insights
What we have learned building and running production systems — written down so it is useful to somebody other than us.
- Vector Search
- Retrieval
Vector retrieval in production: when a vector database earns its cost
A vector database is an architectural decision, not an automatic upgrade to search. If you can't name exactly what you're missing from plain text search, you probably don't need o…
8 min read
- API Design
- Type Safety
Typed contracts between backend and client: why shared schemas cut integration bugs
"It worked in Postman" is a sentence nearly every team has said at least once, right before discovering the real problem was somewhere else. A typed contract between backend and c…
9 min read
- Trust
- News Systems
Scoring news trustworthiness: designing a system that doesn't claim to be neutral
A trust label is an editorial decision encoded in software, not a measurement. Every design choice downstream of that fact — from data model to what you show the reader — depends…
9 min read
- AI Infrastructure
- Model Routing
One model is a single point of failure: routing between hosted and local AI models
Wiring a feature to a single model provider looks like the simple choice — until that provider gets slow, changes price, or fails on one kind of input. Routing across models isn't…
8 min read
- Real-Time Systems
- Reliability
Designing for failure: real-time software that assumes the network will drop
Most real-time systems are designed for the happy path and treat network loss as the exception. In the live-call product we built, disconnection is the rule — and that assumption…
8 min read
- Privacy
- Data
Privacy by architecture: the data you never collect can't leak
The best privacy policy isn't code that deletes data — it's a schema that never made room for it. Data minimization is an architectural decision, not a legal clause.
8 min read
- Operations
- Architecture
Why the on-call rotation matters more than the initial architecture
A system "designed to scale" and a system "designed to be woken up for" are not the same thing. The difference is only visible to whoever actually answers the page.
8 min read
- Observability
- Grafana
Observability that isn't decoration: dashboards people actually open during an incident
Every company has a metrics store and a stack of dashboards. The real question isn't whether you have observability — it's which screen you actually open during a real incident.
8 min read
- Speech AI
- Latency
Live speech translation: where model speed fights speech quality
An ASR-to-MT-to-TTS pipeline adds latency at every stage. In the live multilingual meetings product we built, the real decision wasn't how to make latency zero, but where in the p…
9 min read
- AI Evaluation
- Quality
Why evaluating AI output isn't like testing software
Traditional tests have one correct answer and an assert that checks it. Evaluating a language model's output means judging a range of acceptable answers that keeps shifting undern…
8 min read
- API
- AI Agents
Documenting an API for AI agents is not documenting it for humans
Ambiguity a human developer can shrug off and route around is, for an agent building a tool call, a generation error. The two audiences need different documentation.
8 min read
- Mobile Engineering
- Flutter
Cross-platform vs. native mobile: an engineering decision, not a marketing one
The "one codebase" pitch hides where native actually breaks even. The real question isn't which framework to pick, but which subsystems in your specific product force a native esc…
9 min read
- PostgreSQL
- Observability
- Latency
The p99 that was a cron job
For three weeks one endpoint’s p99 went to nine seconds every night at 02:10 and recovered on its own by 02:40. Nobody was awake to see it and the dashboard averaged it away. The fix was one word of SQL; the finding was the three weeks.
3 min read
- WebRTC
- TURN
What actually connects: TURN, and the networks that fight it
Every WebRTC tutorial ends at “and then ICE finds a path”. On the networks our users are actually on, it often does not. Three years of running a meetings product here, and the co…
3 min read
- SQLite
- Operations
SQLite in production, on purpose
We run several real sites on a single SQLite file, and we would do it again — but not for the reasons people usually give, and not without the four things that make it survivable.…
3 min read
Have something to build?
Tell us what you are working on. We will tell you honestly whether we are the right team for it.
or email us at hello@larsima.com