PTU accelerator: one catalog of 20 Microsoft and Azure AI solutions to increase useful adoption of customers' existing PTUs and demonstrate the value of that investment. There are no bundles or packages to buy into: teams choose whichever solutions fit their work, verify each solution's readiness and model/API/geography routing, and measure business outcomes alongside PTU utilization and total service cost. This is not a blanket PTU-compatibility or deployment guarantee.
Share that link with nontechnical colleagues. The AI Solutions Hub source provides six problem entry points, twenty product-specific solution guides, a multi-solution shortlist and getting-started guidance without exposing this repository's private reports or environment details. The release source and publication evidence are recorded in website deployment metadata. Local changes, including the 20-entry catalog, do not update the hosted website until deployment.
The same site is also served at its original address, https://blue-beach-0fb8cd70f.5.azurestaticapps.net/, which remains live. Azure generated that hostname randomly and it cannot be renamed; see website operations for why both addresses exist.
Private engineering workspace. The stakeholder website is a separate, deliberately sanitized publication. This is not an official Microsoft commercial SKU, a production certification, or a commitment to support the included accelerators.
| Audience | Read |
|---|---|
| Stakeholders | Open the website, or build the self-contained website under site. |
| Account and solution teams | Commercial recommendation: all twenty original candidates, five priority additions and the two later catalog additions. Its three-bundle packaging is superseded by the single 20-solution catalog (see its 2026-09-30 note). |
| DRDC account team | Private DRDC Research Productivity Bundle: a proposed, research-focused composition and gated pilot plan; not public website content or a verified deployment. |
| Engineers | First-pass evaluation, continuation evaluation, catalog additions 21–22, and upstream adaptations. |
| Contributors | Contribution guide, publication boundaries, and website operations. |
The public-facing working brand is AI Solutions Hub, not an official Microsoft marketplace or commercial SKU. The catalog is curated, not an exhaustive list of Microsoft products.
Choose the mix that puts existing PTUs to useful work. Customers can assess one, several or all 20 starting points; deployment remains subject to each solution's readiness, permissions and PTU compatibility. The portfolio is one catalog, not an all-or-nothing package or 20 ready-to-install applications. Three starter pilots (Trusted answers, Document intake and Code modernization) each pair a bounded workflow with ownership/maintenance gates, outcome and quality measures, and a printable pilot brief. CWYD is the first knowledge-pilot recommendation; Content Processing is the document-workflow priority; engineering requires an active backlog and an accountable maintenance owner. Optional extensions are not automatically added to a shortlist. The shared acceptance gates lead to an expand, repair or stop decision. Track repeat use, accepted work, PTU utilization and headroom, time saved, quality and total service cost against a baseline to support investment and renewal decisions. Increased utilization should come from useful work, not artificial traffic; shared services are not included in PTU capacity.
Choose, deploy, adopt, expand. The program works with customers and central AI teams to select a portfolio, arrange approved code access, and either support customer-led deployment or scope end-to-end help through configuration, evaluation, handover and onboarding additional teams. Delivery, funding and operational support are agreed explicitly. SpecSuite is private: access or an approved code handoff can be coordinated with its owner; this does not upgrade its historical evaluation status.
The website's How we help page explains customer-led and assisted delivery, responsibilities and proposed deliverables. A shortlist-based implementation brief collects open solution gates and working-session prompts for local printing or PDF sharing; full technical guides remain a separate print option. It submits no data, books no engagement and does not authorize deployment. Detailed preparation, optional pilots and selection research remain available as secondary guidance.
Use-case ideas reframes the five planned workflows as unbuilt concepts for customer scoping discussions, not a promised release roadmap. Each describes intended users, a useful output, an evaluation approach and a related starting point with explicit implementation gaps. Customers can bring other needs; no priority, delivery date, funding or PTU compatibility is promised. Concepts remain separate from the solution shortlist.
Catalog descriptions explain the input, workflow, useful output and an illustrative scenario for every entry, rather than relying on product-category one-liners. The local visual redesign adds Microsoft branding in the header, switchable workflow illustrations, challenge-led discovery and a three-solution comparison. Quick-save cards and a persistent in-page planning workspace keep the unlimited catalog shortlist separate from comparison; both reset on reload. It uses compact illustrated cards for discovery, with the full explanations and runbooks inside each guide. Original workflow graphics are labeled illustrative; supplied Microsoft logos provide restrained attribution, not blanket product ownership or endorsement. All artwork is embedded for offline use. SpecSuite's supplied product description covers code-to-spec, connected knowledge graphs and spec-to-code for understanding, improving and modernizing existing software. Cross-language intent is not verified support for every language or framework; the package and both directions of the workflow need a scoped evaluation.
Customer catalog curation (23 September 2026): the public offering now contains 20 starting points. Harbinger (historical ID 20) is removed because its owner, package and capabilities are unresolved. StepFly (historical ID 18) is removed because the research prototype did not reach the intended root-cause outcome; its own evaluation excludes it from launch. These are stronger exclusions than simply requiring integration work or providing platform/engineering enablement. Both remain in the private historical inventory and research backlog, not the customer shortlist. Existing reports, evidence and upstream adaptations are unchanged; remaining entries retain their IDs, source pins and readiness limits. Reconsideration requires confirmed provenance for Harbinger and reproducible diagnostic success for StepFly, followed by an approved assessment.
The 20 September 2026 public-source selection review covers 15 current catalog entries and 39 legacy pages without adding them as new validated products. DKM, Modernize and the deployment foundation now show prominent upstream maintenance notices. Their existing package pins and historical selected results are retained; later source review does not establish new functional success. The dated commercial addendum records the updated recommendation.
Visitors choose a problem, compare outcomes and intended users, and open a full solution dossier with product-specific uses, actual workflows, component/connection diagrams, ingestion, deployment prerequisites and a scoped Microsoft specialist engagement. Public sources are pinned where a package is established; platform references and owner-led proposals are distinguished. Shareable section links replace modal dialogs. Visitors can shortlist multiple candidates and print a selected plan. Missing repositories, failed outcomes and custom-build requirements remain documented rather than becoming misleading deployment buttons.
One catalog, ordered by customer priority (30 September 2026). The earlier
three-area grouping (engineering, knowledge and procurement) has been retired.
The site now presents a single list of 20 solutions, ordered by what customers
most often ask about first (customerOrder in site/src/content.mjs). The
first six are highlighted as editorial top picks: Enterprise Knowledge (Chat With
Your Data), Content Processing, Document Knowledge Mining, RFP & contract review,
Modernize and SpecSuite. Top picks are a starting suggestion, not a
readiness, popularity or PTU-compatibility claim. Visitors filter by business
problem instead of area.
Customer-facing cards emphasize value rather than lab status badges or readiness filters. Substantive limitations and the unchanged evaluation context live in each solution's deployment notes. Some selected workflows passed bounded tests; others remain partial, inspected only, blocked, or proposed. Dossiers replace synthetic mockups with descriptions of the real interface and workflow. Source inspection is not a functional test; proposed components and unresolved packages remain explicit.
The DRDC composition is a separate private research proposal. It does not publish meeting-derived information, assert customer endorsement, or authorize deployment. The two supplied AI summaries repeat the same material and require customer validation rather than being treated as independent corroboration.
Current interpretation: the commercial recommendation qualifies and corrects some earlier conclusions. In particular, Voice Live documents a customer-PTU BYOM path, but the lab did not test that path. Do not reuse historical categorical claims without those qualifications.
Historical report filenames, upstream revisions, repository URL and hosting resource names are retained for provenance. Rebranding the experience does not rewrite past evidence or rename infrastructure.
| Path | Purpose |
|---|---|
site/ |
Public-safe stakeholder experience, source, build and tests. Only its generated dist/ directory is deployed. |
reports/, PTU-Bundle-*.md |
Internal evaluation reports, recommendations, cost methodology and runbooks. |
scripts/ |
Evaluation harnesses, guards, deployment helpers and publication checks. Many historical scripts retain MCAPS-specific paths/settings. |
infra/ |
Lab infrastructure definitions and the separate Free-tier stakeholder website template. |
evidence/ |
Reviewed private lab evidence. Presence here is not permission to publish it to the website. |
test-data/, tests/ |
Synthetic fixtures and existing evaluation checks. |
adaptations/ |
Pinned upstream provenance, evaluation patches and authored source overlays instead of vendored dependency trees. |
docs/ |
Publication policy, hosting metadata and operating instructions. |
- The site is informational: no model API, inference credentials, live lab controls or paid-model calls.
- Historical model allowances are closed. Cloning this repository does not authorize restarting inference, increasing quota, purchasing PTUs or deploying the lab.
- Preserve protected existing applications. Use the lab guard and explicit approvals for lab mutations.
- Synthetic MCAPS tests are not a DND deployment or accreditation. Private DNS and identity designs must be adapted to the approved customer landing zone.
- GitHub privacy does not make credential storage acceptable. Never commit keys, SAS tokens, passwords, personal records or original internal briefing inputs.
- Existing Word/PDF briefing inputs, generated Word exports, caches and operational logs remain local and are intentionally not part of the initial publication. Markdown reports are the reviewed source record.
- Accelerators, their service dependencies and individual model deployments have distinct costs and licenses. Purchasing PTUs does not make the rest of the stack free.
Follow site/README.md for build and test commands. The production artifact is self-contained and contains only curated public-safe content.
GitHub Actions validates changes and deploys the site on main changes to the site or deployment workflow. Historical scripts are not automatically executed in CI. Infrastructure provisioning is a separate manual operation; site publishing cannot create or start model-consuming applications.
Six starter issues define the next implementation priorities. Historical functional gaps and approval-gated pilot work remain open; completing the website and private research proposal does not make the twenty catalog entries production-ready or authorize further lab execution. Dependency-update pull requests remain subject to review; they are not automatically merged.
This is a private working collection, not a relicensing of Microsoft or Azure-Samples code. Preserve upstream licenses and notices recorded with each adaptation. Do not redistribute original internal inputs or imply support rights from source availability. No blanket open-source license is granted for the collection.