Andres Kepler — I build software that works in the real world
WHO · 1 / 8
Andres Kepler
I build software that works in the real world.
Hands-on software engineer, systems architect, engineering leader and independent builder.
- Production technology since 1995 — over thirty years.
- Python as my primary language since 2012: production software, not just scripting.
- Extensive PHP before that. Go learned by shipping it at Bolt.
- Java, Haskell, C# and Python coursework at Estonian IT College (no degree). BA, Business Administration and Management, EBS, 2026.
- I live in Estonia, and I'm open to relocating to Portugal for the right opportunity.
Andres Kepler
I build software that works in the real world.
Hands-on software engineer, systems architect, engineering leader and independent builder.
- Production technology since 1995 — over thirty years.
- Python as my primary language since 2012: production software, not just scripting.
- Extensive PHP before that. Go learned by shipping it at Bolt.
- Java, Haskell, C# and Python coursework at Estonian IT College (no degree). BA, Business Administration and Management, EBS, 2026.
- I live in Estonia, and I'm open to relocating to Portugal for the right opportunity.
Monitoring is how I make the invisible argue back.
Not a Grafana keyword — the difference between guessing why a system is slow and knowing.
- Instrumentation turns a hidden assumption into a checkable fact.
- The appeal is the investigation — design, implement, validate — not the dashboard.
- Good monitoring changes a decision. That is the test.
Ask why. Instrument. Find the constraint. Build. Prove it.
Creativity and flexibility, disciplined by evidence — and I stay responsible for what ships.
- 01Ask why
- 02Instrument and observe
- 03Find the constraint
- 04Assess risk and trade-offs
- 05Write and build
- 06Test in reality
- 07Measure
- 08Leave people and systems better off
AI speeds up the implementation today. I remain responsible for the architecture, the review, the tests and the operational result.
Employment — Bolt, 2022–Oct 2026
Keeping a distributed SQL fleet observable — and fixing the tools upstream.
Database Reliability Engineer, Bolt, 2022–Oct 2026. ~19 MySQL/TiDB clusters in DBRE scope — not every one part of every migration.
Monitoring work, as separate steps
- Checkmk→toPrometheusmonitoring migration
- Prometheus→toVictoriaMetricsTiDB metrics, a separate migration
- vmagent→toMimirdual-write into central storage
Python operational tooling, and the Go db-migrator.
Merged upstream at PingCAP — written in Go, learned at Bolt
Role-level background only. No confidential Bolt architecture, incident data, dashboards or customer information.
MoiToi.TECH: where I own the outcome end to end.
My engineering practice since 2019. A long-standing paying client and a product of my own — kept strictly apart.
Zoovet
Client systemPaying client — mine since 1998, through MoiToi.TECH since 2019
I modernise and operate the platform and its infrastructure. I am the person they call.
MerVare
Live pilotMy own product, not a client
I designed the data model and the booking state machine, built both products and the hardware bridge, and operate the live pilot.
loksa.camp: built for people and agents from day one.
A Baltic rental marketplace designed for people, fleets and agents.
- Agents are first-class users: an MCP server, an OpenAPI specification and llms.txt.
- Inventory, booking and real host economics, with a fleet view.
- No traction or revenue claimed yet.
Explore freely. Name the risk before taking it.
What could break, who it hurts, how we back out, and what is still unknown.
Protect the shared system, keep local truth
- Risk
- Central Mimir ingestion came under pressure. The TiDB dual-write stream was the part we controlled.
- Consequence
- Degraded central ingestion is a problem for every team that depends on it, not only ours.
- Mitigation
- Suspend the stream temporarily and rate-limit at the sender. Local VictoriaMetrics keeps the full TiDB telemetry.
- Still unknown
- The central view of TiDB is incomplete until the stream is restored — a known, visible gap.
Validate before claiming traction
- Risk
- Home-berthed boats were being counted as visiting guests in Mervare's demand estimate.
- Consequence
- A business case resting on demand that was not there.
- Mitigation
- The sizing was corrected downward and the published conclusion rewritten.
- Still unknown
- The booking fee alone is a small business. Whether marina software carries the model is still unproven.
Would we work well together?
Curiosity, creativity, freedom, independence, flexibility — shown through choices, not badges.
Curiosity
Checked public AIS tracking against one harbour's own records before trusting the demand story.
Creativity
A Go/ESP32 bridge into a Victron Cerbo GX, so a berth's occupancy can come from a sensor, not a form.
Freedom
An independent practice since 2019, choosing problems I can own end to end.
Independence
Learned Go by shipping it, then took the fixes upstream to PingCAP instead of waiting.
Flexibility
An EBS business degree, five years part-time, alongside senior engineering and live ventures.
I bring most value through hands-on ownership of an ambiguous, difficult problem — not by ticking tickets. If that sounds like the problem you have, let's have a calm conversation about it.