Product updates, events and compliance insights from the Tech Tune AI team.
OSCAR now works directly with Sphinx-Needs, the docs-as-code requirements tool used by many automotive and embedded teams. It reads requirements, specifications, hazards and test cases together with their IDs and links, so every finding points to the exact need. It also writes back: each finding is stored as a new need, linked to the requirement or hazard it concerns, so the gap sits in the same documentation your engineers already review.
Any finding can now become a Jira issue with one click, with the standard clause, the evidence and the suggested fix attached. Assessment gaps land on the same board as all other engineering work. Close the issue, rerun OSCAR, and the next report shows the gap as closed, giving your assessor a complete trail from finding to fix.
OSCAR now checks robot projects against the current editions of the robot-specific standards: ISO 10218-1:2025 and ISO 10218-2:2025 for industrial robots and robot applications, and ISO 3691-4:2023 for driverless industrial trucks and mobile robots. Together with ISO 13849-1:2023 and IEC 62061:2021, which have been live since July, this covers the core of what robot makers and integrators need ahead of the EU Machinery Regulation (EU) 2023/1230, which applies from 20 January 2027. We are following ISO 25785-1 for dynamically stable mobile robots, including humanoids, while it is being drafted.
Our new product walkthrough on the homepage shows the four steps from upload to a prioritised gap report: upload the evidence as it is, review which standard and work products were detected, run the assessment, and open the report. It also shows Thorough mode, where two models check every requirement and their verdicts are reconciled.
At Bits & Pretzels in Munich, our founder Sam Khorshed pitched OSCAR and met founders, investors and industry experts from automotive, health and deep tech. Many conversations came back to the same point: safety and security compliance is becoming a bottleneck as products get more software and more autonomy. Thank you to UnternehmerTUM for the support, and to everyone who stopped by for a coffee and a pretzel.
From Model Year 2027, connected vehicles can no longer be sold in the US if the software behind connectivity or automated driving has a "sufficient nexus" to China or Russia; hardware follows from Model Year 2030, and manufacturers must file annual Declarations of Conformity. Compliance is no longer only about ISO standards and safety cases: "where was this designed, and who touched the code" now matters just as much. Most Tier-1 and Tier-2 supply chains were not built to answer that quickly, and with real automotive development timelines, 2027 is close. By Sam Khorshed, first published on LinkedIn.
In May 2026, EU lawmakers agreed to move the AI Act's high-risk deadline for AI in safety components of regulated products, including vehicles, from August 2026 to August 2028. That is not a reason to relax but a chance to do it properly instead of bolting AI Act compliance onto ISO 26262, SOTIF and ISO 21434 processes at the last minute. The AI Act does not replace existing standards; it adds a governance layer on top: data quality, transparency, human oversight and post-market monitoring. Traceability between engineering documentation and AI-specific requirements should be built into the process now, not retrofitted in 2028. By Sam Khorshed, first published on LinkedIn.
Tech Tune AI has been selected for the Creation Track of the UnternehmerTUM Incubator's Fall 2026 batch in Munich. Over three months we are sharpening OSCAR's problem-solution fit with mentors, workshops and a community of founders, talking to potential customers in automotive, robotics and rail, and preparing our first investment story.
On 30 June and 1 July 2026 we were at GITEX AI Europe at Messe Berlin. In pre-booked meetings we showed the latest version of OSCAR, built for ASPICE, functional safety, SOTIF and cybersecurity, and talked with engineering and compliance teams about where AI can take work off their plates and where human expertise stays essential.
Software-defined vehicles have broken the classic development lifecycle. Software used to be frozen at Start of Production; today a vehicle receives over-the-air updates throughout its life, and every update changes its attack surface. If an update is not continuously checked for cybersecurity compliance, it can fail UN R155/R156 audits and turn a software change into a safety and regulatory problem. That is why we built OSCAR to check compliance continuously, release by release, rather than once at SOP.
OSCAR's compliance engine now extends across two new safety-critical verticals. On the railway side, we integrated EN 50128:2011+A2:2020 and its successor EN 50716:2023 into our standards ontology, covering software for railway control and protection systems across the full SIL spectrum, and giving teams a clear migration path as the industry transitions between the two norms. On the machinery side, OSCAR now supports IEC 62061 and ISO 13849-1:2023, enabling automated assessment of safety-related control systems against both the SIL- and PL-based methodologies. As with our automotive knowledge base, every clause mapping and traceability rule was validated by domain compliance experts before release assessment outputs are expert-proven, not just model-generated. Railway and machinery teams can now run gap analyses, trace safety requirements to specific standard clauses, and generate audit-ready evidence packs in the same air-gapped deployment our automotive partners already rely on.
To support international teams, we rolled out an edge-storage multilingual translation system, allowing users to seamlessly toggle the interface between English, German, and Simplified Chinese without losing deep technical context. Additionally, we overhauled our intake architecture, deploying a seamless, background-fetch request system. Teams exploring Automotive Pilots or Railway/Defence Sandboxes can now submit technical inquiries friction-free, triggering immediate internal routing.
Compliance overhead is a massive hidden cost in engineering. To help management visualize this, we developed and integrated an interactive Engineering Budget Recovery Calculator directly into our platform. Based on our internal OEM cost-modelling, users can now accurately project reclaimed engineering hours and the reduction in external assessor fees. Alongside this, we deployed a significant UI overhaul to make navigating OSCAR's deployment options and sample reports more intuitive.
As pilot users began uploading massive system architecture documents, we realized standard context windows were insufficient. We upgraded our domain-trained encoder stack to support a 32,000 token context window. Furthermore, we finalized our Stage 06 output layer. OSCAR now generates heavily structured, audit-ready PDF reports and machine-readable JSON exports, drastically reducing the time needed to compile external assessor evidence packs.
A massive update to our standards ontology. OSCAR’s knowledge base now fully incorporates all 12 parts of ISO 26262:2018, as well as ISO 21448 (SOTIF). To support this, we upgraded our ingestion pipeline to handle unstructured data, integrating Apache Tika and advanced OCR to parse complex PDF and DOCX exports. The engine can now trace ASIL rationale and flag missing functional safety requirements directly to specific standard clauses.
Listening to strict IT security requirements from our Tier-1 and OEM pilot partners, we completely overhauled OSCAR’s deployment infrastructure. As of this month, OSCAR can be deployed entirely on-premise behind corporate firewalls. We ensured that the inference engine and vector database operate completely locally, meaning zero outbound network calls are made during artifact assessment. Proprietary safety goals and architectures never leave the client's perimeter.
We officially locked in the core RAG (Retrieval-Augmented Generation) architecture for the OSCAR engine. Our primary focus was eliminating the manual mapping of systems engineering artifacts. In this release, we successfully deployed our custom parsers for SysML and ReqIF formats. By mapping these directly against our newly verified ASPICE PAM 3.1 ontology, the engine can now automatically detect gaps in base practices and output capability level scores deterministically, with zero generic LLM hallucination.