Ga naar inhoud
CODT Technologies logoCODT Technologies
Glossary

Every term we use, defined in plain English.

This glossary defines the software and AI terms that come up in our proposals, services and case studies — from AI agents and RAG to nDSG and WebRTC. Every definition is written to stand on its own, and every entry is grounded in something CODT Technologies has actually shipped, with a link to the service or case study where the term lives in production.

29 terms · A–Z · grounded in shipped work

  • Adaptive bitrate streaming (ABR)

    Adaptive bitrate streaming (ABR) is a video-delivery technique that encodes one stream into a ladder of quality levels and lets each viewer's player switch between them mid-playback, so picture quality tracks the viewer's real bandwidth instead of freezing when the connection dips.

    We build ABR pipelines as part of our audio and video streaming service: live ingest over RTMP, SRT or WebRTC, adaptive bitrate transcoding, and per-title encoding so a talking-head clip does not get the same bitrate as fast-moving sport. Perceived quality is checked with VMAF scores. It holds up at scale. The streaming backbone we built for Bizzabo delivers live events to 50K+ peak concurrent viewers with stream start under 1.5 seconds, and RushTix’s ticketed comedy shows hold a 2.4-second median join time with a rebuffer ratio under 0.4%. Those numbers depend on tuning the bitrate ladder to the real audience and their devices.

    Explore our streaming platforms service

  • AI agent

    An AI agent is software that uses a large language model as a reasoning engine to pursue a goal: it breaks the task into steps, calls tools such as APIs, databases and search, observes the results and adjusts — repeating until the job is done or a human needs to step in. Where a chatbot answers questions, an agent takes actions on systems.

    AI agent engineering is one of our core services. Agents get deterministic guardrails (tool-level allowlists, approval gates on sensitive actions, scope-limited credentials), a full audit log of every tool call, and behavioural eval suites that run on every deploy. New agents work alongside people for four to six weeks before they act alone, and per-task cost and latency budgets stop runaway loops before they burn money. Customer-facing agents are built against explicit task-completion targets, and the guardrails are designed so an unauthorised destructive action cannot run in production. A single-purpose internal agent usually takes four to six weeks to build; a customer-facing agent with several tools, approval flows and observability takes ten to sixteen weeks. Every project starts with a paid one-week scoping spike.

    Explore our AI agents service

  • AI voice agent

    An AI voice agent is an AI system that holds real telephone conversations: it listens, speaks and decides in real time. Calls that would otherwise wait for a human are answered, qualified or resolved by software.

    CODT Technologies built and runs one in production: LeadTrack AI, a multi-tenant voice-agent SaaS for an Australian client. We engineered the voice pipeline, telephony integration, per-tenant qualification scripts, intent scoring, tenant isolation and call analytics. The agent phones every new inbound lead within 30 seconds, while the prospect is still at their screen, holds a natural conversation the prospect can interrupt, and routes high-intent prospects to human reps with the transcript and context attached. The platform has completed more than 100,000 calls in production, and calling that fast produced a +38% conversion lift. All of its data stays in AWS Sydney.

    Read the LeadTrack AI case study

  • Answer engine optimisation (AEO)

    Answer engine optimisation (AEO) is the practice of structuring web content so AI answer engines such as ChatGPT, Perplexity and Google’s AI Overviews can extract, cite and accurately reproduce it when users ask questions, instead of only ranking it in a list of links.

    AEO is part of our web app development service. Schema.org structured data (Organization, FAQ, Article, Breadcrumb), clear question-led headings and clean semantic HTML give answer engines something reliable to quote, and the same foundation serves classic SEO. Sites are built to explicit budgets: a Lighthouse SEO score of 100 and Core Web Vitals of 90 or better. We use the same method on our own site. Service pages open with short definition paragraphs, FAQ answers sit in the server-rendered HTML even when collapsed, and structured data mirrors the visible content.

    Explore our web development service

  • Buddy-punching

    Buddy-punching is a form of time theft in which one employee clocks in or out on behalf of an absent colleague, inflating recorded hours and quietly corrupting the payroll data a business pays out against.

    We tackled this for FeelEat, a Swiss food operator, with TapTime, an attendance app that ties each clock-in to the site and the person. Staff tap an NFC tag fixed at the site, scan a QR code or confirm biometrically, and the record goes straight into the ERP that runs payroll. Device binding and verification make it impractical to clock in for a colleague, and a verified clock-in takes about a second. Across FeelEat’s kitchens, logistics and retail sites, buddy-punching dropped by about 95%, attendance accuracy reached 99.5%, and payroll preparation became roughly 80% faster because verified hours no longer had to be retyped. The same record feeds FeelEat’s companion employees app, where payroll accuracy is 99.8%.

    Read the TapTime case study

  • Core Web Vitals

    Core Web Vitals are Google’s user-experience metrics for web pages: how fast the main content paints (Largest Contentful Paint), how quickly the page responds to interaction, and how much the layout shifts while loading (Cumulative Layout Shift). They influence search ranking as well as user patience.

    Our web development service treats the vitals as an engineering budget. LCP under 2.5 seconds and CLS under 0.1 are targets from the first build, checked on every pull request in CI. The work is unglamorous: images in AVIF or WebP at the right sizes with explicit dimensions, non-critical JavaScript deferred, self-hosted fonts with font-display swap, and lazy-loading below the fold. Framework choice matters as well. Astro suits content-heavy sites that should ship almost no JavaScript; Next.js is for pages with real interactivity. Sites are built to Lighthouse Performance of 90 or above and a Lighthouse SEO score of 100.

    Explore our web development service

  • CRM (customer relationship management)

    A CRM (customer relationship management) system is the software of record for a company’s prospects and customers: contacts, deals, conversations and pipeline. Sales and support teams work from one shared history instead of private inboxes and memory.

    We usually integrate with the CRM you already have. Our ERP and business automation service ships production integrations with Salesforce and HubSpot as well as accounting platforms: two-way sync with stable external ID mappings, idempotent updates, a clear system of record per field so conflicts resolve the same way every time, and reconciliation reports that show anything that didn’t match. On the AI side, LeadTrack AI, the multi-tenant calling platform we built, hands qualified leads to human reps with the call transcript and intent score attached. Because we build each agent for the engagement, CRM and telephony integrations are scoped against your actual stack during discovery.

    Explore our business automation service

  • Data residency

    Data residency is the requirement that data be stored and processed within a specific geographic jurisdiction, making the physical location of your servers a legal and contractual matter rather than a purely technical choice.

    We decide residency at the start of an engagement, as part of the architecture. European clients can ask for EU or Swiss data residency, and their engagements are GDPR- and nDSG-aware by default; UAE work is designed with the PDPL in view. We already run residency in production: LeadTrack AI, the voice-agent SaaS built for an Australian client, keeps its data in AWS Sydney and was designed against the Privacy Act 1988 and the Australian Privacy Principles. The Trust Center lists data residency next to NDAs as standard on-request terms, and our cost guides name audit-trail and residency requirements as items scoped into the fixed quote during the discovery sprint, before the build starts.

    Visit the Trust Center

  • Demand forecasting

    Demand forecasting is the use of statistical or machine-learning models to predict future demand from sales history, seasonality and live operational signals, so that inventory, staffing and purchasing are planned against a number instead of a guess.

    We built a demand-forecasting platform for FeelEat that runs at 98% accuracy in production, measured continuously against what actually sold. The models were trained on FeelEat’s sales history plus seasonality, calendar effects and weather, and they forecast per site and per dish, the level at which ordering decisions are made. Errors shrank in both directions: stockouts fell 58% and excess stock fell 41%. The rollout mattered as much as the model. It ran in shadow mode next to human plans until its results earned trust, then went live site by site. The model itself took weeks; the data plumbing, drift monitoring and the planning views people open at 8 a.m. took months.

    Read the forecasting case study

  • Design system

    A design system is a product team’s single source of truth for interface decisions: design tokens for colour, type and spacing, a documented component library, and usage rules that keep screens consistent as the product and the team grow. A UI kit is a folder of styled components; a design system adds the rules that keep it coherent.

    Our UI/UX design service hands over design systems that work in code. Tokens arrive implemented in Tailwind and CSS variables, the component library is mirrored in Storybook so designers and engineers share one source, and accessibility is designed in to WCAG 2.2 AA: colour contrast, keyboard navigation, screen-reader labels and focus states. Designers stay on through the build to review the implementation and fix what does not match the design.

    Explore our UI/UX design service

  • Discovery sprint

    A discovery sprint is a short, fixed-fee scoping engagement, typically one to two weeks, in which a senior team maps a software project before any build begins. It ends in a written scope and a priced proposal rather than an open-ended estimate.

    Every CODT Technologies engagement opens with one, led by a senior architect. The deliverables are concrete: a written scope, a clickable prototype of the riskiest flows, an architecture sketch with key decisions recorded as ADRs, a fixed quote, and an explicit “won’t build” list. Nothing moves forward until that document is signed. You own it, so taking it to another firm to build is a legitimate outcome. Our pricing page publishes indicative bands so you can orient yourself, and the sprint turns a band into a fixed price, with integrations, compliance requirements and the v1 cut-line named before the quote.

    Read the app cost guide

  • Edge computing (IoT edge)

    Edge computing runs processing on or near the device that generates the data, such as an embedded board, a gateway or an in-store machine, instead of shipping everything to a distant cloud. That cuts latency, bandwidth cost and dependence on a connection. In IoT systems this is often called the IoT edge.

    Our IoT solutions service designs hybrid edge-cloud systems by default. Filtering and inference move onto devices when bandwidth is expensive, latency matters or the device must keep working offline; for example, anomaly detection runs on an NVIDIA Jetson so raw video does not have to be streamed upstream. Fleet-wide aggregation and heavy models stay in the cloud. We design for patchy connectivity from the start: devices buffer readings locally and replay them with the original timestamps on reconnect, and critical control logic runs locally, so hardware keeps working in a reduced mode when the link drops. The target for end-to-end latency from device to dashboard is under 800 milliseconds.

    Explore our IoT solutions service

  • ERP (enterprise resource planning)

    An ERP (enterprise resource planning) system is the internal platform a company runs its operations on: orders, inventory, finance, HR and workforce in one auditable system of record. It replaces the spreadsheets, email threads and disconnected tools that pile up as a company grows.

    We build purpose-fit ERPs for companies whose growth has outpaced their tools, rolling them out module by module next to the old process. The longest-running example is FeelEat’s ERP portal, which runs the Swiss food operator’s back office across 200+ corporate clients: orders, staff schedules, employee records, vouchers and event bookings behind one login. It saves around 12 hours per staff member per week, and sales reconciliation that took two days now takes roughly four hours. The same system carries the workforce layer: TapTime’s verified clock-ins sync into it in real time, which made payroll preparation roughly 80% faster. When a standard ERP such as NetSuite, SAP or Odoo fits, we tell you. Custom work is for workflows those packages cannot model without permanent technical debt.

    Read the ERP Portal case study

  • Fixed-price engagement

    A fixed-price engagement is a software contract in which the cost is agreed in writing before the build starts, moving delivery risk from the client's budget onto the vendor's scoping discipline.

    We quote fixed prices on purpose, and we are clear about where the model stops working. The pricing page publishes indicative bands by project shape. A paid discovery sprint then turns the band into a fixed quote in writing before any build begins, because a price set before the scope is known would only price a guess. For open-ended product builds that run for months, fixing the price of changing requirements tends to reward corner-cutting. Those builds run time-and-materials against a banded estimate with quarterly re-forecasting, billed monthly against the 40-hour minimum on the pricing page. What stays fixed is the discovery sprint itself and any well-defined slice of work.

    See the published pricing bands

  • Follow-the-sun development

    Follow-the-sun development is a delivery model in which teams in different time zones hand work to one another, so a project keeps moving outside any single office’s working hours.

    CODT Technologies does not use this model. Our engineering is in one place, Gurugram, India, and each development team and its manager work in the client’s time zone, so no work is handed between sites. Client-facing business contacts for European work are based in Switzerland, and collaboration runs in English, French and German. Founded in Gurugram in 2017, we now serve clients in 10+ countries, including FeelEat in Switzerland for 9+ years and LeadTrack AI, delivered for an Australian client with data in AWS Sydney. Whatever your time zone, a senior engineer replies to enquiries within one business day.

    See the dedicated team model

  • GDPR

    The GDPR (General Data Protection Regulation) is the European Union's data-protection law, governing how organisations collect, process, store and delete the personal data of people in the EU — and its obligations reach any company serving European users, wherever that company is based.

    Our engagements are GDPR-aware by default, with privacy-first architecture from the first commit, and we do not claim a certification. EU data residency is available on request, and we have built this way for European clients for nearly a decade, including 9+ years on FeelEat’s Swiss platform. Country-specific laws sit on the same baseline: Swiss engagements are nDSG-aware, UAE work is designed with the PDPL in view, and Australian builds are designed against the Privacy Act 1988 and the APPs.

    Visit the Trust Center

  • hreflang

    hreflang is an HTML attribute that tells search engines which language and regional versions of a page exist, so a French-speaking searcher lands on the French page instead of the English one. Without it, the language versions of a multilingual site can compete with each other in search.

    Internationalisation, hreflang and locale-based routing are standard in our web development service. Multilingual builds get a CMS structure where translators work per locale without breaking the layout, plus hreflang tags so Google serves the right language. Our advice: machine-translated content tends to underperform, so the i18n plumbing is our job and quality translation is yours or a partner’s. We do this on our own site. codttech.com is served in eleven languages (English, French, German, Italian, Spanish, Dutch, Polish, Portuguese, Arabic, Turkish and Hindi) with locale-based routing and hreflang alternates on every translated page. A page that exists in only one language declares only that one, so no locale points to a translation that does not exist.

    Explore our web development service

  • Human-in-the-loop

    Human-in-the-loop is an automation design pattern in which an AI system carries out routine steps on its own but sends consequential decisions, such as irreversible, financial or customer-facing actions, to a person for approval first.

    In our AI agent projects, this comes down to sorting actions by risk. During scoping, every action an agent can take is ranked by how much damage it could do, and approval gates go only on the irreversible or high-consequence ones: sending money, deleting data, emailing a customer. That keeps human review to a small share of the steps, so the agent stays useful. The approval screen shows the agent’s reasoning and the exact action it proposes, so a person can approve in seconds. New agents also run alongside people for four to six weeks before acting alone, and when a cost or step budget trips, the agent escalates to a human. The aim is that an unauthorised destructive action cannot be executed at all.

    Explore our AI agents service

  • Multi-tenant SaaS

    A multi-tenant SaaS platform serves many customer organisations (tenants) from one codebase and one infrastructure, with each tenant’s data, users and configuration strictly isolated from the others. It is the architecture behind most modern subscription software.

    Multi-tenancy is at the core of our SaaS app development service; how we keep tenants apart is described under tenant isolation below. The pattern runs through the products we have shipped. LeadTrack AI is a multi-tenant voice-agent SaaS where each customer runs its own agents, scripts and pipelines on one codebase. Aviatize serves aviation operators worldwide on isolated tenants of a single platform that manages 1,000+ aircraft. The same architecture powers our white-label model, where partners resell a shared platform under their own brand, with per-tenant configuration and no code changes.

    Explore our SaaS development service

  • MVP (minimum viable product)

    An MVP (minimum viable product) is the smallest version of a product that can be put in front of real users to test whether it is worth building further — a deliberately drawn cut-line, not a rough draft of everything.

    Our mobile app development service puts a focused MVP at 10–14 weeks from kickoff to store submission, and enterprise-grade products at four to seven months. The fixed-scope estimate comes after the discovery sprint, once it can be based on the actual app. Our cost guides treat the cut-line as the biggest budget lever: deciding what a chargeable first version must do, and what can wait, moves the cost more than any technology choice. Drawing that line is the main job of the discovery sprint. Delivery then happens in vertical slices that work end to end (UI, API, database and telemetry), so the MVP is production-quality from day one.

    Explore our mobile development service

  • nDSG (Swiss Federal Act on Data Protection)

    The nDSG (neues Datenschutzgesetz) is Switzerland’s revised Federal Act on Data Protection, in force since September 2023. It modernises Swiss privacy law along lines broadly aligned with, but legally distinct from, the EU’s GDPR.

    Switzerland is a core market for CODT Technologies. Our business-development contacts are based there, client communication runs in German, French or English, and our longest engagement is FeelEat, a Swiss foodtech operator whose platform we have built continuously for 9+ years. Every Swiss engagement is GDPR- and nDSG-aware by default, with privacy-first architecture from the first commit and EU or Swiss data residency on request. The distinction matters because the nDSG is a separate law, and a Swiss company can have duties under both. That is why we name both in our delivery baseline.

    See our Switzerland hub

  • OTA (over-the-air) updates

    An over-the-air (OTA) update delivers new firmware or software to devices remotely, with no technician and no cable. It is the only practical way to maintain a fleet of hardware spread across many physical sites.

    Our IoT service treats the OTA pipeline as safety-critical. Updates roll out in stages (5% → 25% → 100%) with automatic rollback on failure metrics, dual-bank firmware so a failed update cannot brick a device, and signed images so unauthorised firmware cannot load. Device identity is provisioned at manufacturing, so each unit authenticates on first power-up without manual setup, and the pipeline is engineered for update-success rates above 99.5%. FeelEat’s Happy Fridge network runs this pattern: every connected fridge streams stock, temperature and health telemetry and receives updates over the air, so faults are fixed remotely before perishable stock spoils. That is what lets the unmanned shops run 24/7 without staff on site.

    Explore our IoT solutions service

  • RAG (retrieval-augmented generation)

    Retrieval-augmented generation (RAG) grounds a large language model’s answers in your own documents. At question time, relevant passages are retrieved from a knowledge base and given to the model as context, so answers draw on real sources instead of the model’s memory. It is the standard way to reduce hallucinated answers.

    RAG is a core capability of our AI/ML solutions service. In our experience, retrieval quality is mostly an ingestion and chunking problem. Documents are chunked by meaning, metadata filters are added, and hybrid search (BM25 plus dense vectors) runs behind a reranker. Recall@k is measured on a labelled set before anything gets tuned, and the vector database (pgvector, Pinecone or Weaviate) is chosen per project. RAG is also central to how we handle hallucinations: retrieval grounding, required citations and structured outputs, backed by eval harnesses that fail the build when factuality drops. For factual knowledge, retrieval almost always beats fine-tuning, because fine-tuning mainly changes how a model behaves and does a poor job of teaching it facts.

    Explore our AI/ML solutions service

  • Self-service kiosk

    A self-service kiosk is a fixed touch-screen terminal where customers browse, order and pay without staff. It is used everywhere from fast-food ordering to unmanned stores.

    We built the self-service kiosk in FeelEat’s smart-fridge ecosystem: a tablet interface where customers scan, browse and pay in seconds. It shares one live inventory with the fridges and the mobile app, so the screen shows what is actually inside. It runs unattended in production at 99.7% uptime with 99.4% payment success, and the median order takes 42 seconds. Kiosk software is also one of the ready-made foundations in our white-label catalogue, so operators can launch a branded kiosk product within weeks. Together with the connected fridges, the kiosk is the customer-facing end of an unmanned-retail setup that has run in Switzerland for 9+ years.

    Read the Smart Kiosk case study

  • Smart-fridge vending (unmanned retail)

    Smart-fridge vending is unmanned retail through a connected fridge that identifies the customer before the door opens, tracks what leaves each shelf and charges automatically when the door closes — a shop with no staff, open around the clock.

    We built FeelEat’s smart-fridge platform in Switzerland end to end. Called Happy Fridge, it covers the connected fridges, the scan-to-unlock flow, in-app payment, live inventory telemetry and the menu, with nutrition, allergens and ingredient origins shown for each dish. The closed loop is the business model. The door only opens for a known customer who can pay, and the sale settles itself when the door closes, so there is no checkout step to skip; theft fell roughly 90% compared with honour-system fridges. The platform has run for more than 9 years with the same team shipping every release. The app has 40K+ downloads, and the business is now valued at $10M+.

    Read the Happy Fridge case study

  • SOC 2

    SOC 2 is an assurance framework in which an independent auditor attests to a company’s controls for security, availability and confidentiality. Type I covers a point in time and Type II an observation period. Enterprise buyers increasingly require it from SaaS vendors before signing.

    SOC 2 readiness is a named path in our SaaS app development service. We have led several SOC 2 Type II readiness engagements, building the controls and the evidence trail into the product and running a compliance-automation platform such as Vanta or Drata alongside it; the attestation itself comes from an independent CPA firm. Readiness typically takes six to nine months from kickoff. The foundations that make an audit go smoothly are already defaults in our SaaS builds: role-based access control with audit logs, tenant isolation verified by penetration test, secrets in a dedicated manager with rotation, and least-privilege IAM with short-lived credentials. Readiness for SOC 2, GDPR or HIPAA, depending on the industry, is scoped in the discovery sprint so it shapes the architecture from the start.

    Explore our SaaS development service

  • Tenant isolation

    Tenant isolation is the architectural guarantee that no customer of a shared software platform can read or affect another customer's data — enforced in the database and query layer, not just promised in the contract.

    Our default for SaaS builds is a single database with row-level security and strict tenant_id propagation through every query. An integration test suite fails any cross-tenant read, so a regression cannot slip through, and third-party penetration tests check the isolation from outside. Enterprise contracts that need physical separation get a per-tenant schema or database instead. Our production platforms depend on it. Aviatize serves flight schools, clubs and commercial operators worldwide on isolated tenants of one codebase, managing 1,000+ aircraft; LeadTrack AI keeps each customer’s agents, scripts and pipelines separate on a shared voice platform across 100K+ calls; and the white-label model relies on it, since each reseller’s customers get their own isolated tenants.

    Read the Aviatize case study

  • WebRTC

    WebRTC (Web Real-Time Communication) is the open standard that lets browsers and apps exchange audio and video with sub-second latency — the technology behind in-browser calls and truly interactive live streams, where even a few seconds of delay would break the experience.

    In our streaming projects, WebRTC is used on purpose and only where it pays off. RTMP stays the universal, encoder-friendly baseline for ingest, SRT handles contribution over unreliable networks, and WebRTC comes in when the use case needs conversational latency, such as interactive auctions, betting or two-way shows, because it is much harder to scale. Delivery follows the same logic: standard HLS runs at 15–30 seconds of delay, Low-Latency HLS at 2–5, and WebRTC under a second, with cost and buffer resilience getting worse at each step down. WebRTC is part of the shipped stacks for Bizzabo’s event streaming backbone (50K+ peak concurrent viewers, stream starts under 1.5 seconds) and for RushTix’s live comedy platform, which holds a 2.4-second median join and a rebuffer ratio under 0.4%.

    Explore our streaming platforms service

  • White-label software

    White-label software is a proven product built by one company and rebranded by another to sell as its own — the buyer gets speed and a hardened codebase; the end customer sees only the buyer's brand.

    White-label is a separate offering of ours with a published catalogue: self-service kiosk software, employee clock-in, telemedicine patient portals and food-tech ordering. Each can be rebranded end to end (colours, typography, logos, copy, locale, currency), with app-store submission under the partner’s brand. Launches typically take four to eight weeks, at roughly 30–60% of the cost of a comparable custom build, because the engine’s maintenance cost is shared by every partner on the platform. There are two ownership routes: a licence, where core upgrades and security patches keep flowing to all tenants, or a full source-code transfer. Every engagement also includes a fork path. If a business outgrows the shared foundation, we take a clean fork and develop it as a custom build, with no lock-in penalty.

    Explore our white-label service
Keep exploring
AI agent developmentThe FeelEat ecosystemLeadTrack AI case studyApp development cost guideSwitzerland hubTrust Center
Klaar om te bouwen

Een probleem dat het waard is om
goed op te lossen?

Vertel ons over uw product, uw planning en uw randvoorwaarden. We reageren binnen één werkdag met een eerlijke inschatting van fit, scope en het juiste team ervoor.

Plan een kennismaking
Gesprek inplannenChat via WhatsAppMail ons direct