Introduction
Executive Summary
Vision
Build the best "AI in Engineering" agentic solution that turns any pile of artifacts into a trusted peer assistant for engineering and product management teams.
Test
Target Users and Customer Segments
Our goal is to support the following knowledge workers:
| Persona | Description | Primary Needs |
|---|---|---|
| Engineer | Internal R&D | Design decisions based on engineering guidelines and processes. Depend on accurate capture of requirements and architecture decisions. Need routinely to perform verification tasks from design documents, diagrams and code, back to the requirements that drove such design decisions. |
| Research Scientist | Academic or industrial researcher in STEM fields | Need to consume vast amounts of literature to understand new research directions, compare methods and exploration hypotheses. |
| Enterprise user | Business professional roles | Structure work around internal processes and traceable Line of Business (LoB) tools. Make business decisions distilling information from others within or across organization boundaries. Depend on both information consistency and explainability of forecasts down to individual numbers. |
| Graduate Student / Lifelong Learners | Student taking a course or working towards a thesis / paper review | Course-specific structured explanations, canonical references, pedagogical support, tutoring assistance, exams preparation. |
Out of the these segments, our primary target for the initial release of the product is the engineering community in industries that are safety or mission-critical (e.g., telecom, aerospace, automotive, medical devices, finance). We believe we have identified a major outage in the coverage of existing AI solutions in this space and the more complex the engineering, the better our value proposition of what we call Deep Evidence Agents (DEA) dedicated to help engineers better their deliverables from deep evidence-grounded research and requirements tracing tasks.
High-Level Objectives
Common denominators in the life of all engineering efforts are:
- How they handle change.
- How they ground hypotheses and conclusions on evidence.
No matter the engineering job function, new discoveries change previous hypotheses, new customer feedback change previous product requirements, new market trends change previous forecasts, pricing decisions change development and costs of goods targets. All engineers reconcile newly received information and most business or technical conclusions are typically drawn in meetings where new information is organically introduced and reviewed.
It is less common today for engineers to use AI agents as peer assistants and almost unheard of to delegate to them decision making. Despite the ability of AI agents to process large amounts of information quickly, the challenge remains that humans-in-the-loop need to be able to trust the conclusions drawn by these agents. Trust, like in human relationships, is difficult to gain and very easy to loose. There will be trust when the agent explains its reasoning with grounded evidence that ultimately reduces the risk of a decision based on what the agent produced. On the other hand, trust disappears in a second when AI agents deployed in mission-critical industries hallucinate evidence in such a way that the human-in-the-loop is mislead during the review of their answers. Conservatism is the main culture in engineering organizations where safety, reliability and compliance are non-negotiable.
In the following we outline our high level objectives that capture how change and reconciliation is managed in our system in a way that there is always a grounded evidence presented to humans-in-the-loop that ultimately make decisions. We drew inspiration from decades of our own R&D experience in safety and mission-critical telecom systems where change management and traceability to evidence are fundamental pillars of any engineering process.
TODO: This table is incomplete and needs to be finalized based on internal discussions.
| ID | Objective | Type | Success Indicator |
|---|---|---|---|
| OBJ-1 | Increase trust through transparent, verifiable provenance | User | Top 10 performance in all cases where evidence grounding is benchmarked. |