QA 4-18-25.xlsx
XLSX spreadsheet 32 KB Posted
- Solicitation number
- HHM402-23-SC-0002-0001
- Issued by
-
Defense Intelligence Agency
About this file
This document is a Questions and Answers (Q&A) file for the Defense Intelligence Agency's (DIA) NeedipeDIA Commercial Solutions Opening (CSO) Call #CSO-0025-001, focused on Artificial Intelligence and Machine Learning (AI/ML) for the Task Force SABRE. The solicitation seeks innovative AI/ML prototype projects across three technical gaps: (1) Enterprise Multi-Model Large Language Model Application, (2) Multi-Source Data Retrieval and Pre-Processing, and (3) AI-Enabled Document Generation and Assessment. White papers are due by April 30, 2025, with potential award anticipated in June 2025, and the effort will be conducted on the DIA's JWICS network, primarily using AWS services.
Key details include the solicitation being unrestricted and potentially resulting in a FAR-based contract or Other Transaction Agreement (OTA), with no specific small business set-asides. The project targets revolutionary AI capabilities to provide warfighters with decision advantages, emphasizing solutions that can work across classification levels, integrate with existing systems, and demonstrate rapid prototype potential. Offerors must have or be able to obtain appropriate security clearances and facility accreditations, with a preference for open-source models, flexible architectures, and solutions that can quickly integrate into the DIA's existing technological ecosystem.
View the file
Other files
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
Q&A
| NeedipeDIA CSO Call Question/Inquiry and Response List | ||
| Due to large number of questions, additional responses will be available on 4/22/25 | ||
| # | Question/Inquiry | Response |
| 1 | We are performing independent research on the Combined Synopsis/Solicitation (HHM40223SC00020001). We would like to know if this is a brand new contract OR if there is (was) an incumbent performing these services. If not brand new, could you please provide the current / previous contract number? | NeedipeDIA Commercial Solutions Opening (CSO) HHM402-23-SC-0002 is available as general solicitation, in accordance with 10 USC 3458, to seek for innovative solutions in need areas identified in NeedipeDIA Portal. NeedipeDIA CSO allows "Call" feature that solicits for solution in specific and more detailed need area/technical gaps oulined in each "call". NeedipeDIA CSO Call HHM402-23-SC-0002-0001 is announced, seeking for new solutions for technical gaps identified in call announcement. As NeedipeDIA CSO Call is seeking for new solutions, VACA is planning for new contract/agreements. |
| 2 | Can you clarify if this CSO will include any small business socioeconomic program preferences, such as 8(a), HUBZone, SDVOSB, or WOSB set asides, or if it will remain unrestricted for all business sizes? The CSO doesn’t specify a set-aside, but FAR 19.203 encourages considering socioeconomic programs before unrestricted competitions. This question prompts the officer to disclose any plans for set-asides without being overly specific, covering all relevant programs (8(a), HUBZone, etc.) to broaden the response. | CSO Call will remain unrestricted for all business size as resulting award can be both FAR based contract or non-FAR based award. |
| 3 | Will the forthcoming RFP include a requirement for large business offerors to submit a small business subcontracting plan, as outlined in FAR 19.702, to ensure small business participation in this AI/ML prototype development effort? FAR 19.702 mandates subcontracting plans for contracts over $750,000 (or $1.5 million for construction) if subcontracting opportunities exist. This question directly addresses whether FSIs (likely large businesses) will be required to work with small businesses, focusing on a standard FAR requirement without presuming the outcome. | Depending on threshold and award type, large business offerors may be required to submit a small business subcontracting plan. Offeror seeking award is requied to refer to NeedipeDIA CSO HHM402-23-SC-0002 where it outlines subcontracting plan requirement for FAR based award. |
| 4 | Is there a possibility that the solicitation will incorporate evaluation criteria giving preference to proposals that demonstrate significant small business participation, either as prime contractors or subcontractors, to support DIA’s small business contracting goals? This question explores whether small business involvement will be incentivized through evaluation factors, a common practice in federal acquisitions. It aligns with SBA goals (23% of prime contract dollars to small businesses) and allows the officer to clarify the solicitation’s structure without pushing for a mandate. | There are no overarching evaluation criteria set at this time because different type of awards are anticipated from CSO Call. Depending on award type and offeror's business size, small business participation may be required for award. 10 USC 4022(d)(1) includes requirements to participate in Prototype OTA award. |
| 5 | Can you confirm whether the CSO’s scope and anticipated contract value make it suitable for a partial set-aside or reserve for small businesses, as permitted under FAR 19.502-3 or FAR 19.502-4, to enhance small business opportunities? Partial set-asides or reserves allow portions of a contract to be reserved for small businesses when a total set-aside isn’t feasible. This question subtly probes whether VACA is considering such options, referencing specific FAR sections to show knowledge of the process while seeking clarity on structure. | CSO Call is not limited to FAR based contract; announcement does not include set-aise for small businesses. Scope and anticipated value of each award is not known at this stage; it depends on proposed innovative solutions from offerors. |
| 6 | Will the government provide guidance in the RFP on how offerors should address small business utilization, such as through teaming arrangements or subcontracting, to meet DIA’s enterprise-wide AI/ML objectives? This question indirectly asks about mandates by focusing on guidance for small business involvement. It’s broad enough to elicit details about subcontracting or teaming expectations without assuming a specific requirement, keeping the tone collaborative. | Depending on type of award and threshold, offeror may be required to address small business participation. |
| 7 | Has VACA consulted with the Small Business Administration’s Procurement Center Representative (PCR) or the DIA’s Office of Small Business Programs to assess small business capacity for this CSO, as recommended by FAR 19.201? FAR 19.201 requires agencies to work with SBA and small business offices to maximize small business participation. Asking about consultation shows awareness of the process and prompts the officer to reveal whether small business considerations are active, without demanding a set-aside. | Small Business Program completed review prior to CSO announcement. |
| 8 | I wanted to see if you think there is a contract really coming from this, and if there is, would the buyer be willing to set is aside as a Sole Source 8(a) contract, if it qualifies. | The goal of CSO Call is to solicit for innovative solutions for technical gaps identified in CSO Call. Anticipated number of awards are included in CSO Call. Sole Source 8(a) set-aside is not included in CSO Call because award can be FAR based or non-FAR based award. |
| 9 | Where's coversheet? | NeedipeDIA CSO Call is amended to include coversheet in Attachment. |
| 10 | Are we looking for a specific platform (Azure, AWS, GCP, etc.) where we have to build an AI Application? | AWS or hybrid AWS/on-prem. |
| 11 | Are there specific LLMs (e.g., GPT-4, Claude, open-source) that must be pre-integrated, or is the focus on flexibility to add new models?" | Emphasis on flexibility to add models; updated/new versions of open-source models, custom tuned versions of same, or custom models. |
| 12 | How should the system handle 2,000+ concurrent users? Are there latency benchmarks (e.g., <2s response time)? | No benchmarks established, describe system’s method of scaling to efficiently accommodate variable user loads. |
| 13 | What are the primary data sources (e.g., SQL DBs, S3 buckets, on-prem file shares)? Are they multi-classification (e.g., TS/SCI vs. SBU)? | Multiple sources and classifications; system can either manage them natively or be extendable through custom tools. Primary sources are SQL DB or managed file stores accessible through data owners’ API. |
| 15 | What percentage of document processing should be fully automated vs. requiring manual review? | Cannot quantify, varies by use-case. |
| 16 | Are there legacy systems (e.g., SharePoint, Documentum) that the solution must be integrated with? | Various COTS data stores, system does not have to automatically integrate with every proprietary storage system as long as a straightforward workflow is possible with some HITL. |
| 17 | Are there restrictions on programming languages (e.g., Python vs. Java) or frameworks? | No. |
| 18 | Which capability (e.g., RAG, multi-model support, cross-domain data) is the highest priority? | Highest priority use-case is automated, specialized, domain-specific document generation. Prioritize capabilities accordingly. |
| 19 | Are there existing APIs (REST/gRPC) that the solution must interoperate with? | Yes, primarily REST |
| 20 | Do we have to write only about 1 Technical gap area in this Proposal? | No, write to any of the gaps your solution closes. |
| 21 | Can you confirm whether attachments (e.g., figures, charts, tables) are included in the two-page limit for white papers, or can they be submitted as appendices? | Figures, charts, tables should be inlucded in 2 page white paper. |
| 22 | If an organization plans to submit multiple white papers, should each be sent in a separate email, or can they be bundled in one submission with clearly labeled attachments? | Offeror can choose, preference is to include in a single submission to better envision the solutions. |
| 23 | Reference NeedipeDIA+CSO+HHM402-23-SC-0002+CALL+-+TFS+Amendment+1.pdf, Section 3 Technical Gaps, b. TG-2: |
Is the solution requested a big data solution for an existing AI tool or tools? As written, TG-2 seems to be requesting a capability that is meant to enhance or integrate with an AI tool.
If the solution is meant to be inclusive of AI capabilities, what is the AI gap or AI use case for TG-2?
| TG-2 is a data solution meant to integrate with tools used to close TG-1 and TG-3. These tools do not exists as yet. TG-2 does not have to be inclusive of AI capabilities. The AI use cases for the data gathered and processed using the TG-2 solution are described in TG-1 and TG-3. | ||
| 24 | Reference NeedipeDIA+CSO+HHM402-23-SC-0002+CALL+-+TFS+Amendment+1.pdf, Section 3 Technical Gaps (all TGs): | |
| Does DIA prefer the API solutions to be something that corresponds to their domain service or are APis provided by the Cloud service sufficient to satisfy the criteria? | The API solution built into the cloud service is sufficient. | |
| 25 | End Users and Use Cases: |
Does the Offeror have insight they can share about the general uses cases that they are looking to execute with respect to TG-1 in particular? Can the Offeror provide the estimated of number / scale of users (100+ / 1000+ / greater)? What is the primary intended use case(s) for the LLM? Would the use case fall into one of the following scenarios listed below?
- Text summarization
- Content generation (e.g., war game content, scenario generation, exercise material)
- Retrieval Augmented Generation (RAG)
| - Code generation / coding assistance | TG-1 focuses on an application to house and access current and future LLMs. Use cases include general purpose LLM use through an intuitive UI, including but not limited to general chatbot interaction, document summarization/generation, and API for including into current/future automated workflows. As an enterprise solution we anticipate >2000 concurrent users based on usage statistics from similar tools. For TG-3 the primary use case is using advanced RAG with custom data processing (TG-2) to enable automated formatted document creation and identifying key insights within disparate document types. TG-1 covered above. | |
| 26 | Budget Insight: |
ANTICIPATED FUNDING: Funding for TFS varies by fiscal year (FY) and the number of proposals funded. Funding amounts awarded will depend on the availability of program funds and on the number and applicability of submissions received.
Can the Offeror provide additional insight into the budget behind these TGs? Are
| program funds forecasted / confirmed to be available for awards? | No insights offered. | |
| 27 | Technical Gap Prioritization: |
Can the Offeror provide any prioritization or weighting among the three Technical Gaps: (1) Enterprise Multi-Model Large Language Model Application, (2) Multi- Source Data Retrieval and Pre-Processing, and (3) AI-Enabled Document Generation
| and Assessment? | No specific weighting/prioritization. Intent is to find solutions to each gap and tie them together, can be with one tool from one source or multiple tools from different sources. | |
| 28 | In reference to the Call Announcement (section 3.a, 3b, and 3c), it mentions |
JWICS deployment.
Are all submitter project efforts expected to be developed/provided at the TS (JWICS) level from inception or is there an expectation that some or all the proposed projects may occur on unclassified commercial networks and then pushed up or
| replicated on the TS network? | Intent is for all compute to occur on DIA’s JWICS environment. Data may reside at different classification levels but will transition to JWICS for processing and final storage | |
| 29 | Interfacing with DIA IT Services: |
Can the Offeror specify the preferred or required APIs, platforms, or tools for
| integration into networks and IT services? | API requirement is general purpose REST to interface with various data stores and other tools which have similar API configuration. Solutions are not required to already have the ability to access all data stores out of the box but must be extendable to incorporate those connections. | |
| 30 | Environment Requirements: |
Does the Offeror intend to provide the project environment in JWICS for hosting or can submitters offer environments accredited to process TS/SCI for project hosting
| purposes? | Submitters can host on their own accredited environments as long as users on the DIA JWICS network have no impediments to access. | |
| 31 | “In reference to the Call Announcement on page 4, section 3b (TG-2) there is |
mention of a solution that must retrieve from disparate sources, pre-process based on needs of the relevant analysis or AI tool, …” Is the Offeror seeking to have the company perform the work of tailoring data retrieval and pre-processing workflows during the period of performance (e.g., 6-
| month period).” | Offeror expects a well-functioning proof of concept during the period of performance. Select use-cases (not shared here) will form the basis for this proof of concept to demonstrate the realm of the possible with additional features and data sources planned for PIs beyond the PoP. | |
| 32 | “In reference to the Call Announcement on page 4, section 3c (TG-3) there is |
mention of a solution that has the ability to assess LLM responses based on domain-specific knowledge and provide a feedback mechanism to the LLM based on said assessment” Is the Offeror seeking a turnkey solution for this project or would the Offeror consider low-code, no-code UI frameworks for developing a feedback mechanism
| tailored to the domain-specific knowledge in question? | Low-code, no-code HITL frameworks for feedback using SMEs to add domain knowledge and improve the assessment is preferred to turnkey but difficult to change/improve methods. | |
| 33 | “In reference to the Call Announcement on page 4, section 3c (TG-3) there is |
mention of a solution that has the ability to assess LLM responses based on domain-specific knowledge and provide a feedback mechanism to the LLM based on said assessment” Is the Offeror planning to utilize an existing LLM for proposed solutions or is the expectation that the Submitter include a proposed LLM and infrastructure to host the
| LLM as part of their submission? | The preferred solution is described in TG-1 which would feed into TG-3, possibly through the TG-2 solution. If a TG-1 solution is not included in the submission the submitter can assume a solution exists which has the described capabilities. For context, the current TG-1 MVP which we seek to improve is built on AWS Bedrock. | ||
| 34 | I have been reading the Call Announcement CSO-0025-0001. TG-2: Multi-Source Data Retrieval and Pre-Processing is of particular interest. Will this program require computing power at the TS/SCI level or just access / storage? | Compute is required at TS/SCI along with access/storage. | |
| 35 | Can the Government describe how vendors can be provided access to existing LLMs (as described in TG-1) to ensure full fair and open competition? | Current MVP uses open source models available on AWS Bedrock. Assessment will not include custom models (tuned or otherwise) currently in operation. Assessment will focus on extensibility to include tuned or custom models. | |
| 36 | Can the Government confirm that it will consider the usage of other agencies' models as a service as part of a cost-effective solution? | Will not consider using other organizations MaaS. Will consider using MaaS on an existing IaaS platform (e.g. AWS or Azure) as long as there is no impediment to accessing said MaaS on the DIA JWICS environment. | |
| 37 | Can the Government confirm that facilities are available on site to perform this work, or if it will consider vendor facilities to augment delivery on this effort? | Limited space is available on site for work. Will consider vendor facilities with a preference for vendor w/ SCIF to expedite development. | |
| 38 | Can the Government confirm that the availability of existing cloud computing resources that will be leveraged on this effort? If not, should offerors include cost estimates to provide requisite cloud services in our white paper response? | DIA currently has access to AWS services on JWICS. A sub-set of the full suite of services is available. A rough order of magnitude can include different categories within white paper. | |
| 39 | “TG1 calls for a Multi-model LLM. Was this the intent of the government, or was this intended to call for a multi-modal LLM?” | Multi-Model is correct. The application should have various models available with the user having the ability to toggle which model to use. | |
| 40 | Is utilizing leased SCIF space such as Westway an acceptable way to meet the storage requirements for the purposes of the proposal? If so, is a Statement of Facility Availability from the vendor sufficient to meet requirements? | Leased work space is acceptable for the period of performance. Gov’t also has space available on site. | |
| 41 | If utilizing leased SCIF space such as Westway is not an acceptable way to meet the storage requirements for the purposes of the proposal, is a Statement of Facility Availability from a subcontractor sufficient to meet proposal requirements? | If leased space becomes unavailable and the gov’t space available is not large enough for the team then a subcontractor agreement is acceptable. | |
| 42 | If we plan on integrating COTS licenses, does DIA prefer a roughly equal annual subscription or a larger initial perpetual license with lower maintenance fees in out years? | Between those choices, roughly equal annual costs. If integrating various COTS solutions with fees provide a list of services included in those fees and a detailed sustainment plan to ensure everything stays tied together as an appendix. | |
| 43 | Should we identify vendors that we may purchase hardware systems or software licenses from in the whitepaper as subcontractors or should they (as uncleared entities) be listed separately as ODC vendors? Is the delineation of this line whether their staff is performing cleared work on the contract? | Delineate between cleared/uncleared vendors with a brief description of roles. | |
| 44 | Will DIA allow remotely performed product demonstrations during the vendor engagement period? | Limited virtual demos are acceptable using secure systems, however the vendor or appointed representative must be available for regularly scheduled, in-person interactions on site. | |
| 45 | I wanted to mention we have an OTA already in place that DIA sponsors. I wondered if that was something you would be interested in leveraging for SABRE. | Yes, if solution address technical gaps identified in CSO Call, we invite white papers. | |
| 46 | The Call instructions refer to a cover sheet to be downloaded from the announcement attachments but none is included. What other cover page templates are acceptable? | Thank you for the inquiry. Coversheet is added in CSO Call Attachment. | |
| 47 | I’m reaching out to inquire about the Task Force SABRE (TFS) Artificial Intelligence and Machine Learning (AI/ML) opportunity. Specifically, I’d appreciate any insight you can share on whether this acquisition is anticipated to be an 8(a) or small business set-aside. | 8(a) or small business set-aside is not included in CSO Call because award type includes FAR based and non-FAR based award. TFS invites white paper submission that address technical gaps identified in CSO Call. | |
| 48 | I wanted to see if you think there is a contract really coming from this, and if there is, would the buyer be willing to set is aside as a Sole Source 8(a) contract, if it qualifies. | 8(a) or small business set-aside is not included in CSO Call because award type includes FAR based and non-FAR based award. TFS invites white paper submission that address technical gaps identified in CSO Call. | |
| 49 | Could you please send me a copy of the required coversheet to attach to my submission? | Thank you for the inquiry. Coversheet is added in CSO Call Attachment. | |
| 50 | We would like to know if this is a brand new contract OR if there is (was) an incumbent performing these services. If not brand new, could you please provide the current / previous contract number? | CSO Call is announced with the anticipation of new award. | |
| 51 | If our proposed solution addresses multiple technical gaps (e.g., TG-1 and TG-2), should the Offeror submit separate white papers or as one single integrated submission? | Offeror can choose, preference is to include in a single submission to better envision the solutions. | |
| 52 | Are there specific AI/ML capabilities the Government is already pursuing? | For the purposes of this solicitation, the current MVP which is the basis of comparison is built on AWS Bedrock using built-in open-source models connected to bespoke knowledge bases. | |
| 53 | Can the Government provide more details on the existing platforms, cloud environments, or tools, or on the technology stacks and frameworks of the JWICS environment, so the Offeror can plan to integrate with while designing and prototyping? | Most readily available environment is AWS. Minimal on-prem H100s can become available as well. | |
| 54 | Does the Government have a preferred vendor or environment for hosting LLMs internally (e.g., GPU farms, HPC clusters, or on-prem private cloud)? | Nothing preferred. | |
| 55 | Are there any specific data formats (e.g., text, imagery, structured databases) we should design against, for TG-2 and TG-3? | Large emphasis on text and imagery but also unstructured data such as PowerPoint, excel, html. Output for TG-3 is primarily structured forms/reports/presentations where the structure/format already exists for comparison/testing/improving. | |
| 56 | Are there any restrictions on the source of the models (e.g., open-source vs. proprietary)? | Prefer open-source models. Will consider proprietary, however the gov’t reserves the right to conduct a cost-benefit analysis prior to decision. | |
| 57 | For solutions that incorporate RAG workflows (as mentioned in TG-1 and TG-2), do you have a preferred vector database technology, or are we free to propose the vector database of our choosing? | Free to propose as long as the chosen method is interoperable with all solutions. | |
| 58 | Will domain-specific “knowledge” to guide LLM assessments be provided by Government SMEs, or should the Offeror propose methods to encode such knowledge? | Gov’t provided SMEs and/or document repositories will provide “knowledge”, offeror is responsible for proposing methods to incorporate said knowledge. Examples include but are not limited to document correctness (content and format) and valid insights/assessments/conclusions. | |
| 59 | Will the Government furnish representative datasets (especially for training/validation) at various classification levels, or must the Offeror bring their own data sources for initial proof-of-concept? | Government will provide data/SMEs. Preponderance of training/validation data exist on classified networks. If offeror has previously vetted data sources the gov’t is willing negotiate for their inclusion. | |
| 60 | For TG-2, can the Government clarify which file types and data modalities must be supported across classification levels (e.g., scanned PDFs, Office documents, HTML, email, chat logs, images with embedded text, logs, audio, video, telemetry, geospatial, etc.)? Are there specific modalities that are most operationally important for early AI/ML enablement? | For this solution assume documents of various types with images stored either as a native file type or as a scanned in images stored as either pdf or image file containing text. System should be able to parse geospatial data with the assumption the data contains sufficient data to be understandable, i.e. has CRS data attached, follows ISO 6709, is obviously MGRS/UTM, etc. First priority is formatted documents containing primarily text with some graphics stored as pdf/doc | |
| 61 | For use cases involving Retrieval-Augmented Generation (RAG), is the Government considering adoption of model context optimization standards such as the Model Context Protocol (MCP) to better orchestrate and track how data enters LLM pipelines across models, agents, and use cases? | We are not limited to exploring set standards/methods/protocols as long as said transformations provide the needed capabilities and generate output usable to other solutions. | |
| 62 | Will the Government consider architectural approaches for TG-2 that support modular orchestration of AI-ready data pipelines — allowing dynamic swapping of transformation, chunking, embedding, and enrichment modules to leverage current and future SOTA techniques? | See previous answer. | |
| 63 | Will TG-2 solutions be expected to support secure, high-throughput connections to existing government data stores and file repositories across classification levels? If so, will the Government prioritize experience building reliable, high-fidelity connectors in air-gapped or resource-constrained environments? | Yes to high-throughput to existing data stores. Expect prototype to focus first on readily accessible data stores (report archives, message traffic, structured databases) before emphasizing edge-type connections. | |
| 64 | This appears to be a call order released under an existing vehicle called “NeedipeDIA Commercial Solutions Opening”, HHM402-23-SC-0002. Am I correct in my assumption that I need to be an awardee on the vehicle “NeedipeDIA Commercial Solutions Opening”, HHM402-23-SC-0002 in order to respond to Call# CSO-0025-001? | No, CSO Call is a type of solicitaiton for new awards. You don't have to be an existing awardee from NeedipeDIA CSO HHM402-23-SC-0002. | |
| 65 | Is it possible to get a copy of the white paper example in English? | White paper sample offers demonstration of possible format. Detailed format and content requirements are available in NeedipeDIA CSO and CSO Call. | |
| 66 | In regards to the recent AI solicitation your office released, is it acceptable for a company to submit responses to each of the three requirements using different manufacturers? Or does it need to be one manufacturer for all three requirements? | Offeror can choose, preference is to include in a single submission to better envision the solutions. | |
| 67 | Is the objective to deploy LLMs and associated software tools in the existing JWICs-accessible cloud platform, or in on-premises server(s), or at another location? | Not limited. Quickest solution is DIA JWICS cloud (currently limited to AWS). | |
| 68 | Will there be end users outside DIA (greater IC) consuming these LLM and AI/ML services or data? | Current target user is DIA internal. External users possible but should not be a driving planning factor. | |
| 69 | Does DIA envision multi-domain or multi-classification workflows shared with partner agencies? | For simplicity we are assuming all workflows at lower classifications make available their end products to these TG’s solutions. Solutions do not require ability to access/process/move data between classifications. Similarly, workflows to share end products at various classification levels occur outside these solutions. Features which enable said multi-classification workflows are of interest however. | |
| 70 | Is there a broader plan for hosting or federating this capability beyond DIA’s boundaries? | Not currently, should not be a planning factor. | |
| 71 | Does DIA prefer an on-premises-only solution? Or commercial or hybrid clouds? | No preference. Assessment focuses on capability, usability, sustainability, and cost. While where the application rests impacts each the location itself is not a screening criteria. | |
| 72 | Is there an established, high-performance computing (including GPUs, high memory, and storage capacities) dedicated data center or cloud infrastructure within the DIA JWICS environment that is available for the proposed solutions? If not, should we include cost estimates that include in-prem, cloud, hybrid hardware needed to achieve proposed solutions? | The most accessible infrastructure is AWS. Limited on-prem may be made available depending on the capability and costs. Including a cost estimate is preferred. | |
| 73 | TG-2 specific. 1. Please clarify the approximate scale of “large, geographically separated data holdings”? What is the approximate number of separate data systems/environments envisioned to be included in the scope? What is the scale of data (in petabytes, etc.)? What portion (approximate percentage) of the data might be characterized “unstructured”, if any? | Geographically separated scale is global but digitally accessible, no couriers moving snowballs. Many, many different data holdings which cannot be described here. Vendor can assume data are or can be made accessible on DIA JWICS. Data scale is petabytes however the TG-2 solution should be multi-purpose so not reading/processing/storing petabytes at a single time. The preponderance of the data are unstructured which is why we envision TG-1 solutions working with TG-2 solutions. Simple example workflows are generating documents for an ATO (TG-3 solution but the data processing happens in TG-2) or combining social media data with OpenStreetMap data and police reporting to develop suspects. | |
| 74 | TG-2 specific. Are there thresholds with regard to “high velocity” data (e.g. latency SLA under 30~60 milliseconds sufficient)? Other relevant SLA benchmarks that would be relevant to know and include for this scope? | We have not developed benchmarks or metrics beyond “does it work quickly enough and generate a good data”. However, we do not envision processing streaming data. The description of “high velocity” in this case refers to when taken in aggregate across data stores data are constantly updating. The data for specific use cases are rarely “high velocity” but quick enough and large enough that our non-technical user base struggles to keep up using current workflows. | |
| 75 | TG-2 specific. Is it desired that the solution be ‘zero-record-loss’ e.g. maintain in perpetuity a record of all extracted, transformed, and delivered data states in a secured network storage environment? | Not a requirement based on current policies. Sufficient is the ability to provide a solution to TG-1 or TG-3 enough sourcing data to point a user back to the originating documents for citation. | |
| 76 | TG-2 specific. Please clarify any specifications or requirements regarding “the retrieval and pre-processing” of these data sources, whether it preferred to be done in an On-prem (e.g. ‘IC owned and controlled’) server environment, or in a secure private cloud; Assuming deployed behind an existing firewall/security protocol, etc.? | Must be on or accessible to the DIA JWICS environment. No preference to on-prem or cloud. | |
| 77 | TG-2 specific. Please clarify requirements for “solutions must account for data stores at multiple classification levels”– e.g. Meaning requiring separate name spaces for data at different levels of security classification (?), and therefore authorization and access control management within the solution application itself that governs which agency employees are able to access the data that only matches their own classification requirements? | Simply, the data processing must understand the difference between unclassified and classified with various caveats and carry that information forward with the assumption that the originating data are correctly marked. Intent is enabling a TG-1 or TG-3 to generate responses/documents with the appropriate classification markings and citations to the appropriately cleared user. Using name spaces for various data sources (or the transformed data) is a way along with tying into the greater IC identity control system. | |
| 78 | TG-2 specific. Please clarify what’s meant by “solutions must support fine-grained data entitlements management”? Does that mean, In other words, is the Authorization/Access management governed at the individual application level, or through a separate SSO application managed by the agency? (We are able to provision/match employee roles by feature/function/data access level through our platform, within a secure network.) | Connects to above question. The application does not require it’s own entitlements system, sufficient is the ability to access and apply responses for existing systems within a user’s account. Similarly, the application does not require it’s own authentication system but the ability to access and apply the existing authentication system. | |
| 79 | TG-2 specific. Please clarify if “incorporating a cross-domain solution for moving data between classification levels” means building a specific function within the solution to do this, or if it means incorporating a known and accepted vendor/application to do this? | A built-in CDS is not required but the ability to tie into an existing solution is (e.g. AWS DIODE). The intent is eliminating manual data move processes like DOTS. The application does not have to move the data around at the unclassified level. Vendor may assume unclassified workflows which place the data into the low-side CDS pipes are managed by the user or a separate process. Application should automatically retrieve data from wherever the CDS places it on the high-side. | |
| 80 | TG-2 specific. In regard to RAG implementation, are there particular LLM embedding algorithms which should be incorporated (e.g. intfloat/multilingual-e5-large-instruct)? Additionally, are there particular vector database providers which should be considered (e.g. Open Search, Qdrant)? | No specific algorithm/models/configurations required. Capability over format. | |
| 81 | Does the government require a minimum maturity level for the proposed technologies? The mention of a cost structure and the necessity for rapid integration suggest an expectation of a higher maturity level, such as Technology Readiness Level (TRL) 8 or 9. However, there is also language indicating the potential for a Prototype Other Transaction Authority (OTA) award, which may allow for proposals of less mature solutions. | Less mature solutions are acceptable, TRL is not a screening criteria. However, we are interested in rapid integrations of capabilities. That does not mean a completed monolith. We are interested in prototypes that demonstrate a set of the required capabilities, can rapidly be deployed onto the JWICS environment, and show strong potential for more capabilities. | |
| 82 | Each of the three Needs in the requirement document highlights LLM applications as potential solutions. Does the Government prefer LLMs even if an SLM can provide an effective and less resource-intensive solution? Or are SLMs also allowed? | SLMs are allowed. | |
| 83 | Government states in the requirement document that "rather than issuing another CALL, TFS may retain non-selected proposals from this CALL for potential funding if subsequent funds become available." What is the maximum duration that TFS will retain the non-selected proposals? | Approximately one year. TFS is a short duration effort with the intent to sunset in FY26. If prototype OTA is awarded and if prototype OTA is found successful, production OTA can be awarded without further competition, if need and funding are still available. | |
| 84 | The document reads "Coversheet as downloaded in the Attachments" - there is no other attachment on the SAM.gov posting aside from the Call Announcement. How can vendors can access to the Coversheet template for successful completion? | Coversheet is uploaded in CSO Call announcement in Attachment section. | |
| 85 | Should the sustainment plan include ATO costs? | Yes | |
| 86 | If possible, please identify the envisioned target environment and network. | Solutions will reside on the DIA JWICS network. | |
| 87 | Anticipated Project Phasing – Given the emphasis on successful completion criteria and potential transition to sustainment, should offerors structure their white papers with a base pilot phase and one or more option periods? If so, are there any rough dollar thresholds for the initial pilot phase? | White paper can show progress from base pilot phase and one or more option periods. There are no rough dollar thresholds for the initial pilot phase. | |
| 88 | Sustainment Planning Expectations – Beyond pilot delivery, is the government expecting a 12-month sustainment roadmap including personnel FTE estimates, software licensing, and system refresh costs in the white paper? | Length of sustainment/number of personnel/licensing/refresh cost depends on solution offered. Some company can offer labor heavy with shorter timeline solution whereas another company can offer lean solution with longer timeline. It depends on best solution that industry can offer to tackle technical gap shown in CSO Call. | |
| 89 | Engagement Model – For white papers selected for advancement, will offerors have a chance to meet with technical SMEs to refine approach before the formal technical/cost proposal is due? | Follow-on engagement allows offerro to show/demonstrate capability/solution to requirement owners. If selected for request for prototype proposal, offeror can submit formal technical/cost proposal. | |
| 90 | If an Offeror wishes to respond to all three technical gaps, should the Offeror prepare a single 2-page white paper submission that covers all three technical gaps, or should the Offeror submit separate 2-page white papers for each technical gap? | Offeror can submit single white paper to address all 3 technical gaps. Offeror can also submit separate white papers for each technical gaps. | |
| 91 | The Call Announcement states that the selection process "involves two (2) rounds of evaluation" and references the white paper and proposal stages. However, the Timeline and Offeror Deadline section on Page 6 specifies an additional "follow on engagement" stage that includes a proceed or not-to-proceed request element, in addition to the white paper and full proposal stages. What are the anticipated activities / due outs during the follow-on engagement stage? | CSO allows option for mission owner to directly awarding contract/agreement from white paper. Follow-on engagement is an opportunity for mission owner to seek clarification on white paper and better understand solution offered in white paper. Therefore, request for follow-on engagement is not necessarily a down-selection stage. | |
| 92 | The DIA NeedipeDIA CSO HHM402-23-SC-0002 states on Page 15, Section E.4 that white paper submissions may not exceed two pages in length. However, Call Announcement CSO-0025-0001 states that white papers may not exceed four pages in length. Does the maximum white paper length specified in the Call Announcement (4 pages) supersede the maximum length requirement listed in the overall CSO (2 pages)? | Thank you for the inquiry. CSO Call is amended to update page number from 4 pages to 2 pages. | |
| 93 | Will the instructions for the full proposal stage be any different from those outlined in DIA NeedipeDIA CSO HHM402-23-SC-0002, Section F? If so, when will the Government provide full proposal instructions to Offerors? | Full proposal instruction will be provided to offeror that mission owners expressed interest in moving forward with next evalaution. Proposal instruction is not included in this CSO Call because each solution presents different options to engage in proposal instruction based on business status/term restrictions/etc. | |
| 94 | Would DIA consider extending the 2-page limit for the AI/ML Task Force SABRE - CSO-0025-001 white paper, due on April 30, 2025, to 2 pages for the solution and 1 page for the ROM? The justification for this request is that more space is needed to provide the Government with sufficient level of detail to make the ROMs meaningful. The ROM must include a rough cost estimate with projected labor categories, labor costs, a delivery schedule with milestones, projected material purchase costs, and a breakdown of specific other direct costs (ODC). | ROM doesn't need to demonstrate specific materials/trips written out as there will be opportunity to show detailed breakdown in request for prototype proposal phase of evaluation, if selected. | |
| 95 | Will the Government formally accept an Other Transaction Agreement (OTA) with associated in-kind contributions, and if so, can you confirm the terms under which these contributions will be recognized and applied within the scope of the project? | In-kind contribution depends on different solutions that can be offered. Assessment of contribution will depend on what is offered and how value can be assessed. | |
| 96 | Will the Government formally accept a Cooperative Research and Development Agreement (CRADA), including any associated in-kind contributions, and if so, can you confirm the terms under which these contributions will be recognized and applied within the scope of the project? | CSO Call doesn't exclude CRADA arrangement. In-kind contribution depends on different solutions that can be offered. Assessment of contribution will depend on what is offered and how value can be assessed. | |
| 97 | If our white paper addresses capabilities that could meaningfully support both TG1 and TG3 (e.g., document generation using domain-specific LLM orchestration), is it advisable to submit one integrated white paper and cross-reference the dual relevance, or should we submit separate white papers each focused on a single TG? Can you shed any additional selection criteria that would enable offerers to evaluate when and how best to segment TG1/TG3 gaps when an underlying software platform provides both. | Offeror can choose, preference is to include in a single submission to better envision the solutions. | |
| 98 | Is DIA focused on acquiring tools with known successful past performance in the government? | DIA is looking for innovative solutions for Technical Gaps shown in CSO Calls. Solution doesn't need to be tool that's already proven successful in Government use case. |
&"Courier"&12UNCLASSIFIED
&"Courier"&12UNCLASSIFIED
File details come from the government source that posted it. Updated .