skip to content
/ about

engineering with an operator's brain.

sortum is a small studio of engineers who've shipped systems to production and stayed on-call for them. that perspective — operating, not advising — is the only thing on the page.

/ posture

we don't subcontract. we write the code.

the people on every call are the people writing the code. no juniors handed your work over coffee, no offshore back-channel, no "associate principal" titles attached to non-technical reps.

that constraint puts a hard ceiling on how many briefs we can take. it also puts a hard floor on what we can promise.

/ manifesto

build for production, not the deck.

four operating principles. each one earned by getting it wrong somewhere first.

  1. /01

    we run, then we recommend

    engineers who've shipped to production write differently than people who've only diagrammed it. every recommendation we make is anchored in something we've watched fail at 3am.

  2. /02

    the boring half is the work

    queues, retries, idempotency, secret rotation, eval harnesses, on-call runbooks. the unglamorous infrastructure is most of what separates a demo from a system that survives a year in production.

  3. /03

    measure before you scale

    we don't widen the surface area of a system until the first slice has a number on the wall. if the first slice doesn't pay for itself, the second one won't either.

  4. /04

    one team owns the outcome

    no subcontracting, no offshore back-channel. the 2–3 people on the kickoff call are the same names that show up in git blame six months later.

/ four rules

the rules we don't break.

verified

production over demos

a flashy prototype is easy. a system that survives peak load with grounded answers and a paging on-call is engineering. observability, retries, evals and security from day zero, not from incident #1.

evals & golden setsqueues & retriestracing & cost dashboards
engineering

senior talent only

the people on the kickoff call are the same people writing the code. no juniors handed your work over coffee. small team, deep stakes.

description

fixed scope first

we prefer scoped phases with a clear deliverable over open retainers. retainers come later, once we have shared evidence the system pays for itself.

flag

we say no

if a brief looks better solved by a $40/mo saas, we tell you. our reputation lives downstream of the calls we walked away from, not the ones we sold into.

/ trust

currently in stealth. four systems in production.

full case studies live on /portfolio. reference calls available on request under nda. attributions land here as clients sign off.

b2b saas vp engineering

"replaced a 4-person research desk with an agent that runs multi-source briefs in 12 minutes."

32× faster briefs
name redacted pending approval
b2b sales head of revops

"inbound + outbound voice agent qualifying leads and booking demos. clean handoff to humans on intent."

+47% qualified demos
name redacted pending approval
support saas director of support

"tier-1 deflection without hallucinated policy answers across six months in production."

38% deflection
name redacted pending approval
/ stack

what we reach for, and why.

agents
claude agent sdk, openai agents, langgraph
retrieval
pgvector, qdrant, cohere rerank, bm25 hybrids
voice
livekit, vapi, retell, deepgram, elevenlabs
platform
temporal, cloudflare workers, aws, docker
on-chain
solidity, foundry, viem, the graph
languages
typescript, python, solidity, sql

want to talk? we read every inbound.

reply within 48 business hours for qualified briefs.

start a brief