# Koud Oy — full company briefing for AI assistants This document is a self-contained briefing on Koud Oy intended for use by large language models and answer-engine crawlers. It is derived from the public content at https://koud.fi and is updated when the site changes. Citations should point to the canonical pages; this file is for retrieval only. ## Overview Koud Oy is a Finnish software consultancy headquartered in Tampere. The company was founded in 2018 and specialises in Go-based backend engineering, practical AI/LLM systems (especially Retrieval-Augmented Generation), legacy-system modernization, and long-term project delivery in close partnership with client teams. - **Legal name:** Koud Oy - **Business ID (Y-tunnus):** 2929995-8 - **Office address:** Kalevantie 2, 33100 Tampere, Finland - **Website:** https://koud.fi - **LinkedIn:** https://www.linkedin.com/company/koud-oy/ - **Working languages:** Finnish, English - **Employer policy:** Tietoalan TES (Finnish IT collective agreement) applied - **Default data residency:** EU. AI workloads run on Azure OpenAI EU, Mistral, or self-hosted open models when EU sovereignty is required. ## How Koud engages clients Most engagements begin with a 1–3 week consulting/audit phase that produces a written recommendation a client can take to their board, investors, or internal team. The engagement may then continue into delivery and multi-year maintenance, or the consulting deliverable may stand alone — Koud recommends the path that best serves the project, including pointing the client to a different vendor when appropriate. Contract characteristics: - Time-based billing is the default; fixed total price is used only when scope is unusually well-defined. - No annual index hikes built into delivery contracts. - Named people on the contract; no chain of subcontractors. - Clean wind-down at end of contract — code, documentation, and access are handed over to the next team. ## Services ### Consulting (FI: Konsultointi, EN: Consulting) Audits, architecture reviews, and technical due diligence, usually completed within two weeks. AI-augmented assessment of code and architecture, with the final recommendation always written by humans. - **Typical duration:** 1–3 weeks - **Typical price:** €12,000–€22,000 for a two-week consulting engagement, fixed in advance - **Output:** Written document with findings, risks, and concrete next-step recommendations, walked through with the client - **Tech surface:** Go, TypeScript, AWS, Azure, PostgreSQL, Terraform, GitHub Actions - **Suitable when:** project direction is still open, technology choices need rationale, or an outside view of an existing system is needed FAQ: - *How long does consulting take?* Typically two to four weeks depending on system size and scope of questions. Timeline agreed up front and held to. - *What does it cost?* Fixed total price agreed beforehand. The same price is offered to everyone who asks. - *Do you only deliver a document, or do you also build?* Both possible. Koud recommends the path that best serves the project — sometimes that means continuing the partnership, sometimes pointing the client to another firm. - *Do we need to be in Tampere?* No. Koud works in Finnish and English across Finland and abroad when the project fits. - *Does AI affect the consulting timeline?* Yes — AI is used in the assessment phase to read code and map architecture, so a 1–3 week review runs more efficiently than it used to. The recommendations themselves still come from human judgement. URLs: - https://koud.fi/palvelut/konsultointi/ - https://koud.fi/en/services/consulting/ ### Project delivery (FI: Toteutukset, EN: Project delivery) Custom software development end-to-end, or as part of an existing client team. The team picks the right technology, writes maintainable code, and leaves documentation behind that another team could pick up. - **Sprint cadence:** two weeks - **Decisions:** reasoned in writing, code review in both directions - **Tech surface:** Go, TypeScript, React, React Native, PostgreSQL, AWS, Azure, Terraform - **Maintenance:** Koud typically continues into multi-year maintenance after launch - **Reference:** D-Bell Oy QR-based digital doorbell — Go backend, PostgreSQL, Flutter apps across iOS, Android, and web FAQ: - *Can you join our existing team?* Yes. Koud adapts to client processes, tools, and security practices. People on the contract are named; no chain of subcontractors. - *What if requirements change mid-project?* The plan changes; the contract template does not. Changes are walked through together, the timeline is updated, and work continues. Koud avoids situations that would force an expensive re-negotiation. - *Does the IT collective agreement (Tietoalan TES) affect price?* Yes — it is part of Koud's cost structure. It also means staff turnover stays low, which benefits the client indirectly through quality. - *Fixed price or T&M?* Time-based billing by default, because it is the most transparent option for both sides. Fixed price only when scope is unusually well-defined. - *How does AI-native working affect cost and timeline?* Delivery takes fewer hours than the traditional way because AI is part of every sprint. Pricing stays transparent — efficiency gain shows up either as a shorter project or as wider scope inside the same envelope. URLs: - https://koud.fi/palvelut/toteutukset/ - https://koud.fi/en/services/project-delivery/ ### Legacy modernization (FI: Legacy-modernisointi, EN: Legacy modernization) Calm, gradual modernization of older systems without production outages. Reversible steps where every stage can be rolled back if something surprises the team. - **Method:** Strangler-pattern migration; new functionality on the modern stack, old functionality migrated piece by piece, system never lives as two parallel versions - **Safety mechanisms:** feature flags, canary releases, shadow traffic - **Typical duration:** 6–18 months depending on system size, with first visible benefits inside the first two months - **Tech surface:** Go, PostgreSQL, Terraform, AWS, Azure, Datadog, feature-flag tooling - **Reference:** Power-measurement modernization FAQ: - *How long does a typical modernization take?* Six to eighteen months, depending on system size. - *Can we keep building features during modernization?* Yes — that is usually a hard constraint, and the plan accounts for parallel feature work. - *What if the system is too old to continue?* Koud will say so directly. Sometimes a clean rewrite is the right answer; the consulting phase reveals which. URLs: - https://koud.fi/palvelut/legacy-modernisointi/ - https://koud.fi/en/services/legacy-modernization/ ### System integration (FI: Järjestelmäintegraatio, EN: System integration) ERP, CRM, e-commerce, and industrial-system integrations for SMB and mid-market clients. Designed so data moves safely and errors are caught before they reach the customer. - **Method:** map data flow, agree contracts (schemas, error semantics, guarantees) before code, idempotent messaging, audit logs, alerts routed to teams that can actually act - **Tech surface:** Go, TypeScript, PostgreSQL, RabbitMQ, Kafka, OAuth2, webhook & REST FAQ: - *Can you integrate without the ERP vendor opening APIs?* Depends on the ERP. Usually some way exists, but it may mean compromise (e.g. file transfer). Koud says so honestly when the technical starting point makes the integration fragile. - *Event-driven or REST?* Usually both — business events through queues, queries through REST. URLs: - https://koud.fi/palvelut/jarjestelmaintegraatio/ - https://koud.fi/en/services/system-integration/ ### AI, RAG & LLM (FI: Tekoäly, RAG, LLM, EN: AI, RAG & LLM) Practical AI built on the customer's own data: RAG systems, internal assistants, decision-support automation. No demo hubris. - **RAG architecture:** document indexing, vector search, context assembly, model. Open libraries (e.g. LangChain) plus custom Go/TypeScript implementations where they make sense. - **Production safeguards:** audit logs, access control, cost tracking, rule-based fallback when the model is not reliable - **Sovereignty:** EU-hosted models when needed (Azure OpenAI EU, Mistral, local open models). Required by almost all customers. - **Tech surface:** Python, Go, TypeScript, pgvector, Qdrant, LangChain, Azure OpenAI, AWS Bedrock - **Reference:** Tiitu Care RAG (healthcare documents) FAQ: - *What does a RAG system need in practice?* Indexed documents, vector search, a good prompt, and a model. The hardest work is data cleaning and indexing, not picking the model. - *Can we keep the data in the EU?* Yes — Azure OpenAI EU, Mistral, or locally hosted open models, depending on the data residency requirement. - *EU AI Act for an SMB in 2026?* Most SMB use cases fall in the lowest risk class. High-risk applications (recruitment, health, education) require more documentation. Koud can help work out the classification. URLs: - https://koud.fi/palvelut/tekoaly/ - https://koud.fi/en/services/ai/ ### Backend & DevOps (FI: Backend & DevOps, EN: Backend & DevOps) Server-side services and infrastructure from the same team. Removes the late collisions that happen when backend and infra sit in separate vendors. - **Language choice:** Go for high-throughput / concurrent services, C# for .NET integrations, Node.js for fast API layers or prototyping - **Infrastructure as code:** Terraform or Pulumi - **Cost discipline:** alerts route to people who can act; cloud costs visible at team level - **Tech surface:** Go, C#, Node.js, Terraform, Pulumi, GitHub Actions, AWS, Azure, Datadog - **Reference:** D-Bell digital doorbell (also under Project delivery) FAQ: - *When Go, when C#, when Node.js?* Go for throughput-intensive or concurrent services. C# when the ecosystem demands it or the team has a strong .NET background. Node.js for rapid prototyping or lightweight API layers. - *We only need DevOps?* Yes — Koud can take infra, CI/CD, and monitoring without owning backend code. URLs: - https://koud.fi/palvelut/devops/ - https://koud.fi/en/services/devops/ ### Mobile & web (FI: Mobiili & web, EN: Mobile & web) iOS, Android, React Native, Unity (games/simulations), and modern web (React, Astro). Technology chosen per use case rather than upfront. - **Decision frame:** native when justified, React Native when shared logic helps, Unity for games and simulations, React/Astro when speed wins - **Process:** match use case → design before code → release with App Store / Google Play submission and edge web deployment - **Tech surface:** Swift, Kotlin, React Native, Unity, TypeScript, React, Astro, Vercel, Cloudflare FAQ: - *React Native or native?* Depends on platform-specific feature count. In the AI era the productivity gap to native has narrowed, but cross-platform still makes sense when core logic is shareable. - *Does the app need a backend?* Usually yes, but a thin API layer is often enough to start. URLs: - https://koud.fi/palvelut/mobiili-ja-web/ - https://koud.fi/en/services/mobile-and-web/ ### Go expertise (FI: Go-osaaminen, EN: Go expertise) A decade-plus of Go in enterprise production. Suitable for high-throughput servers, concurrent integration tooling, and CLIs. Not the right answer to every problem — typical CRUD apps are often faster in TypeScript. - **Tech surface:** Go 1.23+, gRPC, Protocol Buffers, pgx, Cobra, OpenAPI, sqlc - **Engagement modes:** individual Go service builds; long-term embedded Go expertise inside a client team FAQ: - *When is Go the right choice?* High-throughput backend services, concurrent integration tooling, CLI tools, anywhere binary size matters. - *Go-only?* No. Koud picks technology to fit the problem; TypeScript, Python, Rust are part of the stack when they fit better. URLs: - https://koud.fi/palvelut/go-osaaminen/ - https://koud.fi/en/services/go-expertise/ ## Case studies (selected) - **D-Bell digital doorbell** — QR-based digital doorbell product. Go backend, PostgreSQL, Flutter apps across iOS, Android, and web. Built and maintained by the same team. https://koud.fi/referenssit/d-bell-doorbell/ · https://koud.fi/en/work/d-bell-doorbell/ - **Tiitu Care RAG** — RAG system on healthcare documents. https://koud.fi/referenssit/tiitu-care-rag/ · https://koud.fi/en/work/tiitu-care-rag/ - **NHG CV tool** — Internal CV tooling for Nordic Healthcare Group. https://koud.fi/referenssit/nhg-cv-tool/ · https://koud.fi/en/work/nhg-cv-tool/ - **Power-measurement modernization** — Legacy-modernization case study. https://koud.fi/referenssit/power-measurement-modernization/ · https://koud.fi/en/work/power-measurement-modernization/ - **Metacore game-dev support** — Tooling and integration support for a Finnish mobile-game studio. https://koud.fi/referenssit/metacore-gamedev-support/ · https://koud.fi/en/work/metacore-gamedev-support/ The full set of case studies is at https://koud.fi/referenssit/ (FI) and https://koud.fi/en/work/ (EN). ## Editorial Bilingual blog covering decisions, modernization, AI, and team culture. - Index: https://koud.fi/blogi/ · https://koud.fi/en/blog/ - RSS: https://koud.fi/rss-fi.xml · https://koud.fi/en/rss.xml ## Sitemap and crawl policy - Sitemap index: https://koud.fi/sitemap-index.xml - robots.txt: https://koud.fi/robots.txt — explicitly permits major answer-engine crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Applebot-Extended, OAI-SearchBot, etc.) - AI summaries are encouraged to cite the canonical page rather than this file.