I_5 Sources Sought Questionnaire Attachment.pdf

Market Research Questionnaire

The purpose of this document is for market research. This document does not commit or obligate the Government in any manner. Please complete all fields and provide documentation to support your responses below. Failure to respond fully to this questionnaire will render your response incomplete and your company’s response will not be considered a viable source. Fill in the form completely and send to Parvinder.s.brar@irs.gov no later than May 5, 2026, 2PM PST.

1a. Service Title: Multimodal AI Investigative Platform

1b. website

Company Information:

2. Company legal name:

3. POC: Email: Phone:

4. UEI#:

5. Cage Code:

6. System for Award Management (SAM) Registered and active?

7. Any SAM exclusions:

7a. NAICS: 7b. Business Size:

7c. Socio-Economic Category (8(a), HUBZone, WOSB, SDVOSB etc.): ______________________________

8. GSA contract# (if applicable):____________________

Past Experience: (Note: past experience is not required to be considered.)

10. List company’s past experience providing this service with the Federal government?

Product Contract # Agency Govt POC Name Phone Email

Note: Additional items can be added on continuation page

Service Information:

11. Price per unit:

12. Desired line item structure, e.g. what is the unit of measure for invoices? (Per hour, per job, per month, per square foot)

13. Please provide pricing information for previous projects:

Size Status Information:

14. The primary purpose of this acquisition is to provide Multimodal AI Investigative Platform:

14a. Of the service identified in the SOW/PWS, which one(s) will your company perform as the prime contractor using its own personnel without subcontracting?

mailto:Parvinder.s.brar@irs.gov

14b. With respect to the service actions identified in the SOW/PWS, which action(s) do you intend on subcontracting or using other than your company’s own personnel?

14c. Of those services that will be performed by subcontractor(s), approximately what percentage will they perform for each of the actions identified?

14d. What is the size status of the intended subcontractor(s)? ________________________

14e. If you intend on subcontracting, please detail how you will be in compliance with the Limitations on Subcontracting.

Technical Questions:

1. Evidentiary Traceability & Legal Defensibility

Describe how your solution establishes and maintains deterministic, immutable linkage between all derived data elements and their originating source artifacts.

Your response shall include:

• A detailed explanation of how source-to-output traceability is preserved at the data element level (e.g., page, timestamp, extraction method).

• Description of how version control and corrections are handled without breaking lineage.

• Sample machine-readable provenance output (e.g., JSON, XML, or structured log).

• Explanation of how this capability operates identically in air-gapped environments.

• Identification of any probabilistic or inferred linkage mechanisms, if applicable, and how they are mitigated for evidentiary use.

2. Deployment Architecture & Government Operational Independence

Provide a detailed technical description of how your solution operates fully within Government-controlled, on-premises and air-gapped environments without reliance on vendor-hosted services.

Your response shall include:

• Architecture diagram showing deployment in:

o Fully disconnected (air-gapped) o Restricted network o Connected environments

• Explanation of how AI/ML and LLM capabilities are deployed locally, including model hosting and update mechanisms.

• Description of how the Government can:

o Configure, maintain, and extend the platform without vendor-embedded personnel o Scale compute resources (including GPU acceleration) without architectural redesign

• Identification of any external dependencies, telemetry, or call-home features (must be explicitly disclosed).

3. Enterprise Throughput, Concurrency & Performance Validation

Demonstrate how your solution meets or exceeds the Government’s enterprise-scale workload requirements.

Your response shall include:

• Documented performance metrics for:

o Concurrent users (target ~2,500 users) o High-volume ingestion (financial documents, multimedia >4GB) o Token processing capacity (≥ 1M tokens)

• Description of how the platform prevents:

o License contention o Processing throttling o Operational backlog

• Real-world or production benchmark data (not theoretical estimates).

• Explanation of scaling model (compute vs licensing vs throughput).

• Identification of any hard limits (users, files, processing, tokens, etc.).

4. Salient Characteristics Compliance Matrix (Required Chart Completion)

Responders shall complete all fields in the matrix below.

Responses shall be specific, technical, and verifiable. Narrative-only responses are not sufficient.

• Compliance must be marked as:

Y = Fully Compliant | P = Partial | N = Not Compliant

• “Partial” responses must include mitigation approach, timeline, and associated risk

• “Custom” indicates capability requires non-native development, integration, or modification

• Failure to complete all fields may result in an unfavorable rating

Matrix

Salient Characteristic Area

Requirement Compliance (Y/P/N)

Description of How Requirement is Met Native / Configured / Custom

1.1 Deployment & Infrastructure

Operates fully on-prem, air-gapped, no SaaS dependency

1.2 Deployment & Infrastructure

Functional parity across connected and disconnected environments

1.3

Infrastructure

Operable and extensible without vendor-embedded staff

1.4

Infrastructure

Supports CI-2 HCI and composable GPU infrastructure (Base Period integration)

1.5

Infrastructure

Scales via Government-controlled compute without external data movement

2.1 Performance & Throughput

Demonstrates sustained enterprise-scale processing capacity

2.2

Throughput

Supports high-volume intake without backlog

2.3

Throughput

Maintains reliability under peak workload conditions

2.4

Throughput

Supports enterprise concurrent users (~2,500 users)

2.5

Throughput

Scalable ingestion aligned to annual intake volumes

2.6

Throughput

Supports ≥1M tokens

3.1 Data Ownership & Portability

Government retains full ownership of all data and outputs

3.2 Data Ownership & Portability

Outputs exportable in non-proprietary formats

3.3

& Portability

No proprietary ontologies restricting use or migration

3.4

& Portability

No vendor lock-in or egress barriers

4.1 Multimodal AI

Processes structured, semi-structured, and unstructured data

4.2 Multimodal AI

Performs OCR, handwriting recognition, extraction, transcription, translation locally

4.3 Multimodal AI

AI/LLM operates fully within Government-controlled

4.4 Multimodal AI

Automated classification and workflow routing

4.5 Multimodal AI Human-in-the-loop validation, override, and audit logging

4.6 Multimodal AI

Full transparency and traceability of AI outputs

5.1 Financial Forensics

Processes financial artifacts (bank statements, checks, etc.)

5.2

Forensics

Supports continuously updated financial format libraries

5.3

Forensics

Automated categorization, classification, reconciliation, anomaly detection

5.4

Forensics

Outputs suitable for evidentiary use (FRE 901 & 702)

6.1 Large File & Media

Processes large media files without segmentation

6.2

Media

Preserves timestamps, metadata, and source continuity

6.3

Media

Parallel processing without workflow throttling

6.4

Media

Multilingual transcription/translation in Government environment

6.5

Media

Supports Chinese dialects (e.g., Cantonese, Shanghainese, Jin)

6.6

Media

Maintains linkage between original and translated content

6.7

Media

Supports ≥1M tokens

7.1 Traceability & Provenance

Deterministic linkage between data and source artifacts

7.2 Traceability & Provenance

Data elements include source ID, page/timestamp, extraction method

7.3

Provenance

Immutable lineage with version-controlled corrections

7.4

Provenance

Direct navigation from output to source evidence

7.5

Provenance

Machine-readable provenance logs exportable

7.6

Provenance

Operates identically in offline/air-gapped

8.1 Governance & Explainability

Supports automated triage and decision-support workflows

8.2

Explainability

Outputs are explainable and attributable to source artifacts

8.3

Explainability

Mandatory human-in-the-loop review and override

8.4

Explainability

Monitors performance drift, bias, degradation

8.5

Explainability

Logs system changes, retraining, and user overrides

9.1 Licensing & Sustainability

Supports enterprise licensing (~2,500 users)

9.2

Sustainability

No throttling during surge investigative operations

9.3

Sustainability

No per-use/per-transaction licensing constraints

9.4

Sustainability

Scales compute without duplicative licensing

9.5

Sustainability

Transparent enterprise pricing model

Failure to demonstrate full compliance (Y) with any requirement that impacts evidentiary traceability, deployment independence, or enterprise scalability may result in an unfavorable rating.

Responses marked “Partial” will be reviewed for risk, timeline, and operational impact, and may be deemed unfavorable if they introduce material risk to Government investigative operations.