iiterate Signals | iiterate Technologies

iiterate at Futury Sommerfest 2025 iiterate showed AI automation and the E4RTH project at Futury Sommerfest 2025 in Frankfurt. News •3 min Pinned 09/15/2025 •Frankfurt

AI-Generated Virtual Hosts at Future Tech Fest 2025 AI-generated video avatars as virtual hosts in online VR: the Volkswagen Customer Center case at FTF 2025. Highlight •4 min Pinned 08/12/2025 •Düsseldorf

iiterate at VivaTech 2025: five signals from the floor Five signals from VivaTech Paris 2025, led by data privacy moving from checkbox to differentiator. Highlight •5 min Pinned 06/16/2025 •Paris

iiterate at DigiMesse 2025, Koblenz At DigiMesse 2025 in Koblenz, iiterate showed AI and AR that cut manual effort across the Mittelstand. News •4 min Pinned 06/11/2025 •Koblenz

Aashwin Shares his Entrepreneurial Journey at TUHH Founder Aashwin Shrivastava on the path from international student to AI founder, back at TUHH. Highlight •4 min Pinned 11/28/2024 •Hamburg-Harburg

iiterate at Digital Demo Day 2023 A year into building in Germany: five founder takeaways from Digital Demo Day 2023 in Dusseldorf. News •5 min Pinned 08/17/2023 •Düsseldorf

Real time as a use case: when the frame budget sets the architecture Games, trading and voice agents share one problem: the decision has to fit a window a token-by-token model cannot hold. KI •6 min 09/21/2026 •Worldwide

Typed decisions in the mid-market: six processes that fit Invoice matching, expense approval, claims triage, regulatory review, security alerts and extraction verification. Six processes and the questions behind them. Methodik •7 min 09/21/2026 •Worldwide

What developers are actually building with Jev A week after launch there are dozens of public projects. The patterns inside them say more than the vendor benchmark. Werkzeuge •8 min 09/20/2026 •Worldwide

A number without a reason A decision model gives you a calibrated probability and nothing that looks like an explanation. That is more honest than the alternative. Methodik •6 min 09/18/2026 •Worldwide

Where a decision model actually goes Five places in a working B2B pipeline where a typed decision replaces a language model call that was never language work. Methodik •6 min 09/17/2026 •Worldwide

Not every decision needs a language model TypeSafe AI released Jev, a model that cannot write a sentence. The interesting part is not the model. It is the split it forces. KI •6 min 09/16/2026 •Worldwide

How large can a 3D model be in the browser? A 3D model in the browser has two budgets: transfer and GPU memory. Textures usually decide whether it still runs smoothly on a customer's phone. Methodology •9 min 09/14/2026 •Adenau

Replacing a legacy system: step by step, not on one cut-over date How to replace a grown legacy system module by module: a router in front of it, an order derived from dependencies, and a parallel run with a comparison. Methodology •7 min 09/14/2026 •Adenau

What makes up an AI assistant's response time An AI assistant's response time is made up of five items. Measuring each one on its own finds faster fixes than new hardware or a different model. AI •8 min 09/14/2026 •Adenau

Why AI features fail in practice: seven stages where errors arise A wrong AI answer almost always has an address. Knowing the seven stages of an AI feature lets you find the fault before anyone swaps the model. Methodology •9 min 09/14/2026 •Adenau

Room check and level of detail: does it fit the room, and when does AR help? Whether an enclosure fits the hall can often be calculated without a camera. When a room check is enough, how detailed the 3D model must be, when AR helps. Methodology •9 min 09/14/2026 •Adenau

Funding for AI projects in 2026: programmes, requirements and the path to an application Which funding programmes are open to AI projects in German SMEs in 2026, what they fund, and how to reach a sound application step by step. Methodology •8 min 09/11/2026 •Germany

iiterate at Slush 2026 in Helsinki On 18 and 19 November 2026, iiterate will be at Slush in Helsinki as part of the German delegation. What Slush is and who should talk to us there. Highlight •3 min 09/11/2026 •Helsinki

Claude Fable 5.1 in practice: what the model carries and what it does not solve One million tokens of context, cache reads at a fortieth of the input price, and a retention rule that decides the architecture in Germany. Tools •8 min 09/10/2026 •Germany

Claude Fable 5.1: cheaper in the agent loop, stricter on data Anthropic's new flagship cuts the cache read price to a quarter and is at the same time a Covered Model excluded from zero data retention. AI •10 min 09/10/2026 •Germany

GPT-6 Astra versus Claude Fable 5.1: an honest comparison Most comparison tables for these two models do not survive scrutiny. This piece shows why, and names the small set of numbers that actually holds. AI •10 min 09/10/2026 •Germany

Datenbaum-Fehler: warum Ihre Definition das Richtige rechnet und das Falsche ausgibt Die Geometrie stimmt, die Zuordnung nicht. Datenbäume sind die häufigste Ursache für parametrische Modelle, die plausibel aussehen und falsch sind. Methodik •8 min 08/08/2026 •Adenau

Vom Grasshopper-Modell zum Web-Konfigurator: die Definition als Dienst Eine parametrische Definition muss nicht im Entwurfswerkzeug bleiben. Sie kann als Dienst laufen und Konfigurator, Angebotsrechnung und Prüfung aus derselben Logik bedienen. Werkzeuge •9 min 08/08/2026 •Adenau

Wann sich ein parametrisches Modell rechnet, und wann nicht Ein Regelwerk zu bauen dauert länger als die erste Variante zu zeichnen. Ab welcher Variantenzahl sich das umkehrt, und die Fälle, in denen es das nie tut. Methodik •8 min 08/08/2026 •Adenau

Was ein RAG-System im Betrieb wirklich kostet Nicht das Modell ist der Kostenblock. Es sind Indexpflege, Evaluation und der Betrieb, und diese drei stehen in keinem Angebot, das nur die Entwicklung beziffert. Methodik •9 min 08/08/2026 •Adenau

Baidu OCR in the stack: why classic text recognition stays alongside visual search OCR is not dead. PaddleOCR-VL delivers searchable text where visual search alone falls short. Tools •2 min 06/25/2026 •Worldwide

Baidu Unlimited-OCR: one long document in a single pass, on-prem Baidu's Unlimited-OCR keeps the KV cache constant and reads long PDFs in a single pass, MIT-licensed and on-prem. Tools •5 min 06/25/2026 •Worldwide

Context engineering: why it matters and how to do it well Context engineering designs everything a model sees, not just the prompt. It is what makes agents reliable. Methodology •3 min 06/25/2026 •Worldwide

F-RAG (RAG-Fusion): how it differs from plain RAG RAG-Fusion runs several query variants and fuses the results with reciprocal rank fusion. Better recall, some drift risk. Methodology •2 min 06/25/2026 •Worldwide

Figma Motion: animation moves into the canvas Figma Motion brings keyframes and a timeline directly into the canvas. For teams, that changes the handoff break. Tools •2 min 06/25/2026 •Worldwide

GLM-5.2, Sakana Fugu, Claude Fable 5: three frontier models, three answers on control Three frontier models, one real question for a B2B buyer: where does your data go, and how much of the stack do you control? AI •4 min 06/25/2026 •Rheinland-Pfalz

Google Omni: announced, shipped, and enterprise-ready are three different things Gemini Omni turns any input into video, but as of mid-2026 there is no API and no EU-residency path for enterprises. News •3 min 06/25/2026 •Worldwide

LangChain, LangGraph, LangSmith, LangFlow: four names, two companies, one decision Three of these four are one converging stack. LangFlow is a different company. Pick by the job, not the brand. Tools •5 min 06/25/2026 •Worldwide

Late interaction explained: why Qdrant and ColQwen build the better knowledge base Late interaction compares every search term against every page excerpt. Qdrant stores that natively. Methodology •2 min 06/25/2026 •Worldwide

ModernVBERT: a small model, on-prem ready ColModernVBERT reaches near-ColPali quality with 250M parameters and fits on modest hardware. Tools •2 min 06/25/2026 •Rhineland-Palatinate

n8n versus Make.com: which automation for which team n8n is self-hostable and developer-friendly, Make.com is faster for no-code. The choice comes down to control and team. Tools •2 min 06/25/2026 •Rhineland-Palatinate

The OCR-free document stack 2026: ColPali, ColQwen, ModernVBERT, and Qdrant Visual retrieval models search the page as an image instead of via OCR. A sober overview. Methodology •3 min 06/25/2026 •Rhineland-Palatinate

From photo to printable part: what image-to-3D can and cannot make Image-to-3D makes a pretty mesh, not a manufacturable part. A printable part needs geometry a slicer trusts: watertight, scaled, tolerant. Methodology •3 min 06/25/2026 •Worldwide

Prompt engineering vs context engineering: not a rebrand, a change in what you optimise Prompt engineering words one instruction. Context engineering designs the whole payload the model sees, under a token budget. Methodology •4 min 06/25/2026 •Worldwide

Query decomposition and the advanced-RAG toolkit: match the method to the failure Query decomposition, HyDE, RAG-Fusion, GraphRAG each fix one failure. The sophisticated 2026 RAG knows when to use none. Methodology •4 min 06/25/2026 •Worldwide

RPA versus AI agents: when classic automation still wins RPA beats AI agents everywhere the process is stable and structured. Reliability against flexibility. Methodology •2 min 06/25/2026 •Rhineland-Palatinate

SAM 3D Body: single image in, a riggable human out, and where it quietly breaks Meta's open SAM 3D Body rebuilds a full human mesh from one photo, runnable in your own pipeline. Pose-plausible, not metric-truthful. Research •3 min 06/25/2026 •Worldwide

The end of OCR? Visual document search and what it changes for the Mittelstand Visual retrieval models find tables and scans where OCR-based RAG fails. What that means in practice. AI •2 min 06/25/2026 •Rhineland-Palatinate

Seedance 2.5: native 30-second AI video, no stitching ByteDance's Seedance 2.5 generates 30 seconds in one pass and takes 50 references. The seam problem shrinks. News •2 min 06/25/2026 •Worldwide

Vibe coding: where it helps, where it bites Vibe coding is a tool, not a method. The line runs between throwaway code and software that ships to production. Methodology •3 min 06/25/2026 •Rhineland-Palatinate

YOLO in 2026: NMS-free detection, a version map that lies, and the licence nobody reads YOLO26 ships NMS-free detection and faster CPU inference, but the highest version number is not the supported one. Tools •4 min 06/25/2026 •Worldwide

Automating business processes with n8n (and where AI comes into play) n8n automates processes self-hosted in the EU. Where AI helps, where it gets fragile, and when real engineering is needed. Tools •4 min 06/19/2026 •Rhineland-Palatinate

GPT Image 2 vs Nano Banana Pro: choosing an image model GPT Image 2 and Nano Banana Pro win different jobs: on-image text, 4K, consistency, or stack fit. How we choose. Tools •4 min 06/19/2026 •Worldwide

Image-to-3D with TRELLIS: 3D assets from a prompt TRELLIS turns one image or a prompt into a GLB 3D asset, MIT-licensed. What it nails, and the cleanup it still needs. Tools •5 min 06/19/2026 •Worldwide

No-code agent builders for the Mittelstand: what they do and where they stop No-code agent builders ship value fast, then stop at governance, evals and scale. When to use one, and when to build. Tools •5 min 06/19/2026 •Rheinland-Pfalz

Voice agent platforms compared: ElevenLabs vs Retell vs Voiceflow ElevenLabs, Retell and Voiceflow solve different voice-agent jobs. How to pick for German support, CX, or dev teams. Tools •4 min 06/19/2026 •Worldwide

Why we built our own ZUGFeRD tool, and whether we should share it We built a lean, self-hosted ZUGFeRD tool for the e-invoicing mandate. Should we share it? Highlight •4 min 06/18/2026 •Germany

Choosing a coding agent: Claude Code, Codex, Cursor, Antigravity, or Opencode Your data situation decides the coding agent, not the benchmark. Does your code leave the building or not? Tools •5 min 06/18/2026 •Germany

Extending agentic harnesses: skills, artifacts, commands and connectors The model gets the headlines. The harness around it is where your leverage and your attack surface both live. Tools •5 min 06/18/2026 •Worldwide

AI solutions for the Mittelstand in Rhineland-Palatinate The RLP Mittelstand already uses AI, mostly informally via US cloud. Sovereign and subsidised is the better path. Methodology •4 min 06/18/2026 •Rhineland-Palatinate

AI video models 2026: Veo 3, Kling, Seedance, and the open Wan Animate In AI video, quality became mandatory in 2026. The B2B tiebreaker is data handling, labelling, and whether the model is open. News •4 min 06/18/2026 •Worldwide

Open-weight model for German B2B: Qwen, Kimi, Nemotron, or Minimax? For open-weight models in German, the license decides before the benchmark does. Three of the four are genuinely free. Tools •5 min 06/18/2026 •Germany

Personal knowledge management with AI: Claude Code and Obsidian Your Obsidian vault stays local. The AI's inference does not. That split is the whole PKM decision. Methodology •5 min 06/18/2026 •Worldwide

The state of AI development in Rhineland-Palatinate Rhineland-Palatinate researches AI at the national level. Reaching the Mittelstand still has to happen. AI •4 min 06/18/2026 •Rhineland-Palatinate

Subquadratic LLMs: cheaper long context for on-prem RAG Subquadratic models lower the cost of long context. A real lever for on-prem RAG, but still young. Research •5 min 06/18/2026 •Germany

API, MCP, or CLI: choosing how to wire AI into your stack API, MCP, and CLI are three layers, not rivals. Pick by how many integrations you run and who has to audit them. Methodology •5 min 06/17/2026 •Worldwide

Building event-scout: an agentic micro-app that runs unattended We built a small tool that scouts B2B events for us and runs on its own. The lesson is the shape, not the events. Methodology •5 min 06/17/2026 •Remagen, Germany

Where a coding agent earns its keep outside engineering Claude Code's real value outside engineering is not code. It is letting marketing and ops encode their own procedures. Methodology •6 min 06/17/2026 •Worldwide

GDPR and AI: what is forbidden, what is conditional, and what you control Under the GDPR, very little about AI is flatly forbidden. Most is conditional, and the conditions are architectural. Methodology •6 min 06/17/2026 •European Union

Sovereign European AI, honestly: Mistral, the rest, and what it means for a German business Europe's AI scene is finally serious. The honest read is more useful than the flag-waving one for a buyer. AI •6 min 06/17/2026 •European Union

EU AI Act 2026: what companies need to implement now The EU AI Act has applied since August 2024 and is tightening in stages. What concretely lands on companies in 2026. Methodology •3 min 06/16/2026 •Germany

Introducing AI in the Mittelstand: a practical roadmap A roadmap for AI in the Mittelstand: start small, measure against the real problem, keep control of data. Methodology •3 min 06/16/2026 •Rhineland-Palatinate

A local LLM in the company: hardware, cost, reality A local LLM needs less hardware than you think. What matters is matching the model size to the task. AI •3 min 06/16/2026 •Rhineland-Palatinate

On-premise vs. cloud LLM: when local AI is the right choice On-prem or cloud is rarely a model question. It is a question of data sovereignty, cost, and control. AI •3 min 06/16/2026 •Rhineland-Palatinate

What is RAG? Retrieval-augmented generation explained for the Mittelstand RAG couples a language model to your own documents, so answers stay verifiable and current. AI •4 min 06/16/2026 •Rhineland-Palatinate

When the best AI model vanished in 72 hours: the Fable 5 lesson Fable 5 topped the benchmarks, then vanished in 72 hours. Why operational control beats rented capability. AI •4 min 06/15/2026 •Worldwide

Graphify: giving AI coding assistants a map Graphify turns a codebase into a queryable knowledge graph: local-first, no code leaves the machine, cheaper agent runs. Tools •4 min 06/15/2026 •Germany

Kimi K2.7 Code: open-weight, coding-first, cheaper tokens Moonshot's Kimi K2.7 Code: open-weight, coding-first, cheaper tokens. The numbers, the caveats, why open matters. AI •4 min 06/15/2026 •Germany

Nemotron 3: NVIDIA's open models built for agents NVIDIA's Nemotron 3 Ultra: 550B open-weights, agent-focused, strong on the index, free day one. A real on-prem option. AI •5 min 06/15/2026 •Germany

NVIDIA Cosmos 3: one open model for physical AI NVIDIA Cosmos 3: one open world model for physical AI, folding vision, world generation, and action into one system. AI •5 min 06/15/2026 •Germany

iiterate Turns 4: Four Years of Iteration and Trust Four years in: trust compounds faster than the work, and sovereignty is engineered, not declared. News •4 min 06/02/2026 •Germany

Our Insights on the Changing Face of AI in 2025 Where AI already delivers value, and why clear expectations decide whether adoption works. News •4 min 10/23/2025 •Online

Grasshopper 3D: parametric design without code Grasshopper turns 3D modeling into parametric, data-driven design, no code. What it is, how it works, where it fits. Tools •3 min 10/23/2024 •Worldwide

Where AI actually earns its place in a business AI earns its place three ways: automating repeat work, surfacing patterns in data, and fitting how you already work. AI •4 min 09/23/2024 •Worldwide

Why UX decides whether software gets used UX is more than usability: user, system, context. Why it decides whether software gets used, with the numbers. Highlight •5 min 06/15/2024 •Worldwide

Retrieval-Augmented Generation, and why it underpins sovereign AI RAG grounds a model's answers in retrieved sources, cutting hallucination and staleness. Why it underpins sovereign AI. AI •5 min 04/01/2024 •Worldwide

iiterate Participates in Digitale Woche Dortmund iiterate in the international program at Digitale Woche Dortmund, with founders from seven countries. Highlight •4 min 10/23/2023 •Dortmund

iiterate receives the Gründungsstipendium Rheinland-Pfalz 2023 iiterate Technologies GmbH is on the published list of 2023 Start.in.RLP recipients. Highlight •1 min 06/23/2023 •Rhineland-Palatinate