Copy of Copy of ImageCapture QnA_answered1.xlsx
XLSX spreadsheet 16 KB Posted
- Attached to
- RFI for FBI Image Capturing
- Solicitation number
- Not on record
About this file
This is a Questions and Answers document from the FBI Image Capture RFI (Request for Information) that addresses vendor inquiries regarding the agency's market research for an innovative Image Capture and Document Processing solution to support IT modernization efforts.
The FBI seeks a comprehensive image capture platform capable of processing 100,000 images daily on average with peak volumes of 500,000 images daily. The solution must support multiple document types including investigative forms, handwritten documents, photographs, and mixed media, with approximately 25-50% of content requiring Handwritten Text Recognition (HTR). Key intelligent document processing capabilities required include sensitive information detection, handwriting recognition, document classification, metadata extraction, and entity extraction with FADGI compliance. The platform must integrate with FBI's Sentinel enterprise case management system and custom internal applications through standards-based interfaces such as REST APIs and web services. The solution must support Microsoft SQL Server 2022 Enterprise Edition and be compatible with existing scanners including IBML Fusion/IT Series, Ricoh fi-7700 models, Opex Falcon+, and Epson equipment currently in use.
Deployment will initially focus on one centralized scanning location (DocLab) with two remote secured locations, supporting 100 concurrent users and approximately 2 scanning workstations per major scanner model. The solution must operate immediately within the FBI's secret enclave without requiring external cloud-based AI services, though FedRAMP High-authorized cloud or hybrid features may be demonstrated as optional enhancements. The FBI requires support for top ten languages including English, Spanish, Chinese, Arabic, Russian, Persian, Korean, German, and Ukrainian, and sensitive data detection capabilities for classified markings (TS/S/C/U), dissemination controls (NOFORN, ORCON, REL TO), SCI compartments, and all PII, with AI-detected sensitive data flagged for human review rather than autonomous decision-making. Four environments are required (Development, Test, Pre-Prod, and Production), and vendors may respond as manufacturers, authorized resellers, systems integrators, or prime contractors, either with integrated single-platform solutions or combinations of core platforms with third-party components for OCR, handwriting recognition, and AI capabilities. Respondents should identify primary pricing drivers and may provide rough-order-of-magnitude cost estimates if available.
View the file
Other files
| File | Type | Posted |
|---|---|---|
| ImageCapture RFI.pdf |
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
Sheet1
| ImageCapture RFI QnA | |||
| Question # | Section / Page # | Question/ Comment | FBI Answer |
| 1 | Native & Extensible Recognition Services (page 4) | Beyond OCR, what Intelligent Document Processing (IDP) capabilities are considered most important (e.g., document classification, metadata extraction, entity extraction, document summarization, intelligent routing, and sensitive information detection)? | Sensitive Information Detection and handwriting recognition capabilities |
| 2 | File Management and Export – Seamless Integration (page 5) | Which enterprise repositories, document management systems, and business applications will the solution be expected to integrate with? | The solution will be expected to integrate with custom-developed internal business applications and any image capture software that would not be replaced by the proposed solution. Main internal applications being used are for the purposes of receiving/barcoding material prior to scanning. Integration should be supported through standards-based interfaces such as REST APIs, web services, message queues, database connectivity, or other documented integration methods. |
| 3 | File Management and Export – Seamless Integration (page 5) | Which enterprise repositories, case management systems, or downstream applications will consume the captured documents and extracted metadata? | Sentinel - FBI's Enterprise-level case management system |
| 4 | Performance and Reliability (page 5) | Approximately how many documents, pages, and batches are processed daily, and what peak processing volumes should the solution support? | 100,000 images daily (average) |
500,000 images daily (peak)
| 5 | Image Capture and Quality Control (page 4) | What types of documents comprise the majority of the workload (e.g., forms, investigative reports, photographs, handwritten documents, oversized documents, mixed media)? | Mostly textual images of various types (investigative and administrative forms). Digitized evidence can vary widely depending on what was seized. Material is often intermingled with photographs, receipts, books, newspapers, magazines, pamphlets, brochures, maps, blueprints, oversized material, etc. |
| 6 | Native & Extensible Recognition Services – Handprinting & Cursive (page 4) | What percentage of the document population contains handwritten content requiring Handwritten Text Recognition (HTR)? | Approximately 25-50%. Often handwritten text is intermingled with typed text. |
| 7 | Metadata and Indexing – Consistent Schema Application (page 4) | Will metadata schemas be standardized across the enterprise, or will individual business areas maintain unique metadata models? | We are striving for FADGI compliance which requires us to capture various descriptive, technical, and administrative metadata associated with the record being digitized. |
| 8 | Customizable Workflow Automation (page 5) | Will workflows extend beyond image capture into downstream investigative, records management, or other business processes? | Yes, we would like the capability to extend our workflows into downstream processes so they can use the information captured as resources for investigative purposes. |
| 9 | RFI Submittal Instructions – Industry Day (page 9) | During Industry Day, will vendors be encouraged to demonstrate innovative capabilities beyond the stated requirements that could improve operational efficiency, document processing, and investigator productivity? | Yes, additional capabilities can be explored after demonstrating the requirements outlined in the RFI. |
| 10 | Scanner Compatibility – Driver Support (page 3) | Is there a preferred scanner driver for performance and compatibility? | Drivers must be compatible with the scanners outlined in the RFI. |
| 11 | Database Management and Database Compatibility – Database Flexibility / Easy Data Access (page 3) | For database flexibility, normally a database administrator would be the direct user. Can you clarify who/what needs the ability to save, find, and update information and what form of access would be used? | Compatibility: Microsoft SQL Server 2022 - Enterprise Edition |
Who: Users (specified by ADGroup), Adhoc Users, and Service Accounts
| 12 | • Cloud and Hybrid Deployment Capabilities (page 5) | |||
| • Security & Compliance (page 6) | Does the FBI’s desired AI capabilities need to operate within the secret enclave or can they be in a secure cloud such as FedRAMP? | Both - Immediately in the secret enclave. Future capabilities for FedRAMP are desired. | ||
| 13 | Cloud and Hybrid Deployment Capabilities (page 6) | If the AI capabilities are on-premises, what kind of access is allowed to update those models? | Ideally, we would want to be able to update the models in-house. | |
| 14 | Cloud and Hybrid Deployment Capabilities – Compute Offloading (page 6) | For the optional compute offloading capability, does the FBI expect the proposed solution to support GPU-accelerated processing for compute-intensive operations (e.g., OCR, classification, AI inference)? If so, are there preferred technologies or frameworks (such as NVIDIA CUDA) that the solution should leverage? | We don't currently leverage dedicated GPUs for our offloading tasks, but we may in the future. | |
| 15 | What products currently make up the existing capture environment? | OpenText Captiva Suite (ScanPlus, Completion, RescanPlus) Opex Software (CertainScan, Edit, Transform, and Monitor), Ricoh (PaperStream Capture), IBML Capture Suite, and custom-coded applications | ||
| 16 | How many scanners are currently being used by model (IBML, OPEX, Ricoh, Epson, BancTek, etc.)? | 2 IBML Fusion 8000 Series |
3 IBML IT 6000 Series 1 Intelliscan XDS 5 (BancTec) ~80 Ricoh fi-7700 2 Opex Falcon+ 10 Epson Perfection V800 Photo 1 Ricoh ScanSnap ix1400ture.
| 17 | What types of documents are expected to be processed by the new solution, along with the estimated annual volume for each (for example: investigative records, forms, evidence, photographs, handwritten notes, fingerprint cards, financial records, emails, legal documents, intelligence reports)? | See #4 |
| 18 | Does the FBI have a preference for AI models provided by: Microsoft, AWS, Google, On-premises LLMs, Vendor-Provided AI | No |
| 19 | If operating within Secret environments, should cloud-based AI services be completely disconnected, requiring all AI processing to occur on-premises? | The solution must function completely within the Secret environment without requiring external cloud-based AI services. Cloud or hybrid features may be demonstrated as optional enhancements but must be FedRAMP High authorized and must not impact the solution's core capabilities if unavailable. |
| 20 | How many environments will the FBI require for the solution (Development, Test, QA, Production, DR, etc.)? | Development, Test, Pre-Prod, and Prod |
| 21 | The RFI states that the proposed software must hold FedRAMP High certification. Does this requirement apply to the entire proposed solution, including software deployed exclusively on-premises within a classified enclave, or only to cloud-hosted components and services? | The FedRAMP High requirement applies only to cloud-hosted components and services. Software deployed exclusively within the on-premises classified enclave is not required to hold FedRAMP High authorization. |
| 22 | The RFI states that it is open to manufacturers. May an authorized reseller, systems integrator, or prime contractor respond on behalf of, or jointly with, the software manufacturer, provided the response includes the manufacturer’s capabilities and commitments? | Yes |
| 23 | May respondents propose an integrated solution consisting of a core image-capture platform and supported third-party components for OCR, handwriting recognition, artificial intelligence, analytics, or FADGI validation, or is the FBI seeking all capabilities from a single software product and manufacturer? | Either |
| 24 | Is the FBI seeking to replace all current image-capture and document-processing applications with one platform, or should the proposed solution coexist and integrate with selected existing systems? | Either |
| 25 | Please provide estimated deployment volumes, including the number of scanning locations, concurrent users, scanning workstations, annual pages or images, peak daily volume, average batch size, and maximum expected batch size. | Scanning Locations: 1 Main Scanning Location (DocLab), 2 remote secured locations, need to be able to scale up |
Concurrent Users: 100 Scanning Workstations:
2 IBML Fusion 8000 Series 3 IBML IT 6000 Series 1 Intelliscan XDS 5 (BancTec) ~80 Ricoh fi-7700 2 Opex Falcon+ 10 Epson Perfection V800 Photo 1 Ricoh ScanSnap ix1400
Annual Images:
2 Contex HD Ultra-X 6090
100,000 images daily (average) 500,000 images daily (peak)
| 26 | Will the system be deployed at one centralized location, across multiple FBI facilities, or in a geographically distributed enterprise environment? | Currently just focused on 1 centralized location. |
| 27 | Please identify the languages and writing systems that must be supported for OCR, handprint recognition, and handwritten-text recognition, along with any required minimum accuracy levels. | Top 10 Languages/Writing: |
English Spanish Simplified Chinese Traditional Chinese Arabic Russian Persian/Farsi Korean German Ukrainian
Our current OCR process has over 70 languages avaliable to us.
28 What categories of sensitive information must the AI/ML capability detect, such as classification markings, personally identifiable information, law-enforcement-sensitive information, grand-jury information, source identities, investigative terms, or other agency-defined categories? Classification level:
TOP SECRET (TS)
SECRET (S)
CONFIDENTIAL (C)
UNCLASSIFIED (U)
Dissemination controls (examples below) NOFORN – Not releasable to foreign nationals.
ORCON – Originator Controlled.
REL TO USA, FVEY (or other countries) – Releasable only to specified foreign partners.
PROPIN, FED ONLY, and other authorized dissemination controls, as applicable.
Control systems / compartments (these are only used when applicable - see examples below) SCI compartments (e.g., SI, TK, HCS) SAP identifiers Other authorized handling caveats.
Special handling markings Such as Controlled Unclassified Information (CUI) categories for unclassified material, or other authorized handling instructions.
This information is displayed in banner lines, portion markings, and classification blocks.
Any/All PII
| 29 | Is AI-based sensitive-data detection intended to provide advisory flags for human review, or is it expected to make authoritative classification, segregation, release, or access-control decisions? | Advisory flags. All AI-detected sensitive data would undergo human review. |
| 30 | Will the FBI establish acceptable precision, recall, false-positive, and false-negative thresholds for automated sensitive-data detection? | Performance thresholds have not yet been established and may be defined during requirements refinement and acceptance testing. Vendors should describe their solution's detection accuracy, evaluation methodology, and ability to configure sensitivity to balance precision and recall. |
| 31 | The RFI states that learning must occur on a secure network through closed user input. Please clarify whether this requires model training or fine-tuning entirely within the classified enclave, or whether approved models may be trained externally and imported into the enclave after security review. | The FBI has not prescribed a specific model development approach. Vendors may propose solutions that leverage externally developed or pre-trained models, provided they can operate securely within the FBI environment. Any use of FBI operational data for model training or fine-tuning should occur within the secure environment unless otherwise authorized. |
| 32 | Because the RFI requests identification of pricing drivers but is not a request for quotation, does the FBI seek only qualitative cost factors, or should respondents provide rough-order-of-magnitude licensing, implementation, hardware, support, and operating-cost ranges? | Respondents should identify the primary pricing drivers and may provide rough-order-of-magnitude cost estimates if available. Detailed pricing is not required as part of this RFI. |
File details come from the government source that posted it. Updated .