DARPA STTR DPA26TZ05-DV003: SPEED DIAL (Scalable Platform for Enterprise Engineering and Deployment towards Mathematics for the Discovery of Algorithms and Architectures)
Below is a brief summary. Please check the full solicitation before applying (link in resources section).
Quick Answer
DPA26TZ05-DV003, known as SPEED DIAL, is a DARPA Direct to Phase II STTR topic seeking a platform that brings AI-driven algorithm discovery out of the research lab and into everyday engineering workflows. The award is worth up to 750,000 dollars over a 12 month base period, with an optional 1,250,000 dollar, 12 month extension, for a total period of performance up to 24 months. Proposals are due September 23. Because this is an STTR topic, a formal research institution partnership is required.
What This Topic Is Looking For
DARPA's DIAL program already proved that AI can autonomously discover novel, high-performance algorithms, for example rediscovering the Kalman Filter using Transformers, rediscovering wavelets using genetic programming, and generating optimal meta-solvers for physics simulations. The problem is that these discovery engines are stuck in research environments. Engineers and scientists cannot yet pull a bespoke, AI-discovered algorithm into their actual workflow before running their standard process.
SPEED DIAL asks for a framework that closes that gap, built through a partnership between a university with deep algorithmic discovery expertise and a company representing the US industrial base. The Phase II work breaks into five tasks: building a unified discovery and deployment platform where engineers can define a problem space, boundary conditions, and hardware constraints to kick off discovery; embedding pervasive discovery engines, such as Transformer-based or genetic programming approaches, directly into the platform so industrial partners can discover bespoke algorithms for their own proprietary data; curating a version-controlled library of discovered algorithms organized by problem class with performance benchmarks; building in-context integration tools that pair a problem description with its algorithmic solution so retrieval is context-aware, and that deploy the algorithm into existing environments like MATLAB, Simulink, COMSOL, or custom C++ without manual code rewriting; and demonstrating the whole framework on at least two distinct defense-relevant problems such as hypersonic vehicle design, submarine acoustic signature analysis, or digital twin modeling.
A key requirement running through the whole topic is interpretability. Discovered algorithms cannot be black boxes. They need to be composed of transparent building blocks that a domain expert can actually understand, verify, and eventually certify.
DP2 Feasibility and Phase II Structure
This topic accepts Direct to Phase II proposals only, so no separate Phase I award will be made. Feasibility documentation needs to show three things already accomplished outside the STTR program: a track record of using methods like Transformers, genetic programming, or reinforcement learning to discover novel algorithms that beat state-of-the-art approaches in domains like time-series analysis, data compression, or partial differential equations; quantitative performance gains, with DARPA citing prior results like a 1000x improvement from AI-discovered wavelets and a 6.35x reduction in iterations for acoustic and sonar solvers as reference points; and evidence that the discovered algorithms are interpretable rather than opaque.
Phase II runs through a set of fixed milestones: a Month 4 system architecture document with an initial library of at least 5 foundational algorithms, a Month 8 prototype of the ambient discovery engine running in the background without interrupting the primary engineering workflow, a Month 12 demonstration of in-context deployment into a standard commercial environment without manual code rewriting, a Month 14 interim demonstration on a defense-relevant problem showing more than 10 percent improvement in computational efficiency, a Month 18 beta release of the full closed-loop platform, and a Month 24 final demonstration on a second major defense application along with the final software release and Phase III transition plan.
Phase III Outlook
On the defense side, DARPA points to faster and more accurate hypersonic vehicle design through better CFD simulation, improved sonar and radar signal processing for target detection, optimization of logistics networks in contested environments, and faster development of digital twins for military systems. Commercially, the same discovery framework applies to advanced manufacturing process optimization, financial modeling for risk analysis and trading, accelerated molecular dynamics simulation for drug discovery, and semiconductor chip layout optimization.
Funding and Timeline
Base award amount: up to 750,000 dollars for a 12 month period of performance.
Option amount: up to 1,250,000 dollars for an additional 12 months, bringing the total period of performance to 24 months.
TABA: DARPA offers up to 6,500 dollars per Phase I project and up to 25,000 dollars for the initial Phase II award.
Proposal deadline: September 23. Technical questions must be submitted by September 16, since DARPA does not answer questions submitted within 7 calendar days of the closing date.
STTR Requirements
Because this is an STTR topic, the proposing small business must partner with a qualifying research institution, and the topic itself is explicitly framed around a university-company partnership: a university bringing deep expertise in algorithmic discovery, paired with a company representing the US industrial base that can embed the tools into a real engineering workflow.
Proposal Format
This topic uses the standard Technical Volume format rather than the white paper and slide deck format. The Technical Volume splits into Part One, Feasibility Documentation, capped at 10 pages, and Part Two, the Technical Proposal, capped at 20 pages. A separate Phase II commercialization strategy section is required and capped at 5 pages, placed as the last section of the Technical Volume and not counted against the main page limit.
All proposals go through DSIP and require the standard seven volumes: cover sheet, technical volume, cost volume, Company Commercialization Report, supporting documents, Fraud Waste and Abuse training, and the Disclosures of Foreign Affiliations webform, which must be completed as a webform and will not be accepted as a PDF upload.
Frequently Asked Questions
What is the deadline for DPA26TZ05-DV003? September 23. Technical questions must be submitted by September 16.
Is this a Phase I or Phase II award? Phase II only. This topic accepts Direct to Phase II proposals exclusively, so proposers must document feasibility already achieved outside the STTR program rather than receiving a separate Phase I award.
Why does this topic require a research institution partner? This is an STTR topic, and the underlying concept is explicitly built around a university and industry partnership, pairing algorithmic discovery expertise from academia with a company able to deploy that work into real engineering practice.
What does interpretability mean for this topic specifically? Discovered algorithms cannot function as black boxes. They need to be built from transparent components that a domain expert who is not an algorithms specialist can understand and eventually verify or certify, which DARPA treats as essential for real-world adoption.
What page limits apply to the proposal? 10 pages for feasibility documentation, 20 pages for the technical proposal, and a separate 5 page commercialization strategy that does not count against those limits.
Is cost sharing required? No. Cost sharing is permitted under this BAA but is not required and will not be used as an evaluation factor.
What kind of team fits this topic best? A university partner with strength in algorithm discovery methods such as genetic programming, reinforcement learning, or Transformer-based approaches, paired with a small business that has real engineering workflow experience and a credible path to embedding the tool into commercial or defense engineering software.
