AI4ODS: How much does standardization amplify AI when engineers analyze measured data? HighQSoft answers that question in a new two-and-a-half-minute film. It follows an AI connected through the MCP Server for ASAM ODS as it investigates racing data, from the first question to the root cause. Every number in the film comes from the data, and every time shown was measured. The result is an ASAM ODS root cause analysis in three prompts.

Watch the film on YouTube: Standardization & AI: Analyzing Racing Data with the ASAM ODS MCP Server

The data comes from many laps driven by three AI pilots on a racing simulator. Each lap carries 560 channels and about 1.7 million values, all stored in ASAM ODS. One pilot, the simulator's own Ingame-AI-Pilot, sets a fastest lap of 1:03.577. The HQS-AI-Pilot is 2.8 seconds slower, the 3rd-party-AI-Pilot 3.7 seconds. The question is simple. Why is the Ingame-AI-Pilot so much faster?

Until now, every question of this kind meant a new script. Export the data, write the analysis, plot the result, and repeat for the next question. The film replaces that loop with three prompts.

Step 1 builds a live lap dashboard. One prompt asks for a dashboard of the fastest lap of each AI. After 7:50 minutes it runs on live ASAM ODS data, with track position, lap traces and a lap comparison. The HQL queries behind it can serve as the base for an application of its own.

Step 2 turns the analysis into reusable Merlin jobs. One prompt asks why two AIs are level and both slower than the game's own AI, and asks for a throttle and brake KPI, a PDF report and jobs to reuse them. Three minutes later two jobs run on the Merlin Analysis Server. The first writes the KPIs back to ASAM ODS for every lap. The second builds a ten page lap report and updates it when a new competitor drives the track. The report shows where the time goes, in the corners.

Step 3 completes the ASAM ODS root cause analysis. One question asks why the game's AI wins in the corners. Within one to three minutes the AI rules out car, tires, fuel and road, which are the same for all three. It then compares lateral grip at 90 to 110 km/h, 1.55 g for the Ingame-AI-Pilot against 1.26 g and 1.28 g for the others, and looks at the tire curve. The answer is a grip factor of 1.2. The game's own AI drives with about 20 percent more tire grip than physics allows, so it cannot be beaten. The root cause was found by asking, not by scripting.

The AI does the thinking, and the MCP Server knows ASAM ODS. The MCP Server connects Claude, ChatGPT or your own agents to your measurements. It turns a question in natural language into HQL, validates the query against the application model and runs it. It works on every ASAM ODS model without setup, so no script stands between the question and the answer, and every answer stays traceable to the measurement it came from.

The standard does the rest. Because the racing data sits in ASAM ODS with its units, channels and structure described, the AI does not have to guess what the data means. That is the point of the film. Standardization and AI reduce time to insights to minutes, for dashboards, applications, new automated jobs and root cause analysis. More on the combination is on the page ASAM ODS and AI.

#AI4ODS #ASAM #ODS #AI #MCP #HQL #Merlin #TestData #HighQSoft

ASAM has published a new interview in its Fun with Standards series, and it opens with the question every test organization is currently asking. Marius Dupuis, CEO of ASAM, asks Dr. Ralf Noerenberg of HighQSoft whether generative AI makes a data standard like ASAM ODS obsolete. The answer runs in the other direction. The standard amplifies what AI can do with test data, and AI amplifies the value of every standardized dataset. The conversation is on ASAM's YouTube channel, and HighQSoft has published a written companion page carrying the same arguments.

Watch the interview on ASAM's YouTube channel. The written version, Standardization and AI: ASAM ODS, is published on highqsoft.com and works through the argument section by section.

Please find the link and additional content here: Standardization and AI: HighQSoft in ASAM's Fun with Standards Series

The starting point is that AI does not remove the need for a data contract. A measurement still has to carry what it means, which channel is which, in what unit, and under which conditions it was recorded. In regulated engineering work, where a result has to stay traceable years after the test that produced it, that contract is what makes an answer checkable rather than merely plausible.

What AI does change is the consumption layer. Conversational interfaces and agents increasingly take the place of the classic rich client, and standardized interfaces are what keep the tool use underneath them deterministic. The layer below is commoditizing at the same time, which moves ASAM ODS from being only a query target toward being the federation layer across data lakes, object stores, and databases.

The interview is direct about what happens without a descriptive layer. An organization that stores measurements in a lake and skips the semantic model has to rebuild that model itself, constrain which tools can read the result, and accept long-term dependence on a single provider. One top-tier OEM has done exactly that at petabyte scale with its own engineering team. For most organizations, adopting the standard costs less than rebuilding it.

The closing argument is about ownership. Open standards keep the meaning of measurement data with the customer rather than with a vendor, which matters most to mid-sized companies that cannot fund a semantic layer of their own. It is also what lets automation and AI-assisted analysis start immediately instead of after a pre-processing project.

*#ASAM #ODS #AI #TestData #Standardization #HighQSoft*

Following the release of Anthropic's newest Claude model, Fable 5, HighQSoft ran a focused experiment. Could an AI agent derive a reusable skill that creates ASAM ODS dashboards on measured data?

The specification was about 20 lines long and contained mostly boundary conditions, namely Streamlit as the dashboard framework, pyHQL for data access, and the HighQSoft MCP Server for ASAM ODS as the discovery layer. The dashboards themselves were a free pass.

The directions were to discover the database, create five dashboards, keep the skill generic by representation type (graph, geo, table), offer features such as sliders and filters based on the discovered application model, and reinforce lessons learned from each step.

The agent ran for 25 minutes with one manual feedback loop. The result on HighQSoft's weather demonstration database was five working dashboards, a Channel Explorer, a Climate Map, Database Statistics, Last Data Imports, and a Measurement Channel Comparison. Every dashboard runs on live HQL queries against the ODS server. No data was exported, copied, or staged.

Two components made the result possible.

• First, the HQL engine accessed through pyHQL. The agent had the complete query language context available and validated every query against the live server before baking it into an app. What ships is what ran.
• Second, version 2 of the MCP Server for ASAM ODS, rebuilt on FastMCP 3. The key design change in version 2 was the addition of two tools, discover_model and find_paths. Together they let the agent resolve the customer-specific application model instead of guessing element and attribute names. The MCP Server simply knows what the database looks like.

The five dashboards are the visible output, but the actual result is the skill that now builds the next ones. Because the skill works from model discovery rather than hard-coded names, it transfers to any ASAM ODS application model. Test engineers describe the view they want in plain language, and the skill composes, validates, and ships a live dashboard from there.

#ASAM #ODS #AI #MCP #HQL #Streamlit #TestData #HighQSoft

The article then shows how HighQSoft's stack already plugs AI into the standard. The MCP Server connects Claude and ChatGPT to ASAM ODS through HQL. Merlin AI extends the same pattern to analysis pipeline composition. MoMa AI, from the AEOPT research project, generates importer mappings from sample data so a new measurement source becomes queryable in hours.

*#ASAM #ODS #AI #TestData #HQL #MCP #HighQSoft*