Compare Business Plans

jay662

TitanCompendiaOverviewDiscussionwithAFRL_STO10NOV21.pdf

Home >Architecture and Design homework help >Compare Business Plans

Compendia DATA PLATFORM Discussion with AFRL/STO

10 Nov 2021

Agenda 1 Introductions

2 Compendia product Overview & Background

3 Demonstration

4 Questions & Answers

5 Next Steps

Who is Titan Technologies?

2000 Averstar, Pacer

Infotec Joins Titan Corporation

2001 BTG Joins Titan

2006 Titan Joins

L-3 Communications

2014 Data Tactics joins L-3 NSS

2016 L-3 NSS Joins

CACI

2017 CACI Divests

Titan Technologies

A legacy OF SUCCESS

Our leadership team has been implementing complex technology solutions for Defense and secure Federal customers for more than 30 years.

TOP SECRET FACILITY CLEARANCE

200009001:2015

Technology Partners

Expert System USA’s Natural Language Processor software (Expert

NLP) is a sophisticated Commercial-off-the-Shelf (COTS) Natural

Language Processing/Understanding (NLP/NLU) software suite

which enables an organization to process textual content and

autogenerate sophisticated metadata to organize, discover, and

explore information and power analytic tools.

With Expert NLP, the data management paradigm shifts from

time-intensive manual tagging and one-dimensional search to

comprehensive auto-generated metadata and rapid and precise

discovery of information. Expert NLP is operational in the

Defense, Intelligence and Law Enforcement communities.

NNData’s Mission is to deliver a data management platform

that supports “Machine Learning for the Masses”. Their

NNCompass product is a complete AI-enabled data

management platform that lets you get a handle on the

most difficult ETL, streaming, data flow, preparation &

transformation tasks faced in the enterprise today.

NNCompass is operational in the Defense, Intelligence and

Enterprise SW communities.

Titan Team

• David Ramirez - Chief Operating Officer

• Michael Edwins – Chief Growth Officer

• Belay Tekalign – VP, Mid-Atlantic Region, LANL/NSRC Program Manager

• Rob Buntz - Director, Digital Business Transformation

• Brian Welde - Compendia Product Manager

• Kevin Collins - Compendia Architect

• Kenny Kawahara - Compendia Engineer

• Gerald Gay – Chief Technology Officer

• Kevin Cousin, PhD – Chief Data Scientist

• Randy Garrett, PhD – Chief Data Officer

• Steve Aberle – Compendia Engineer

• Giuseppe Strafforello – Chief Technology Officer

• Emily Pace – Principal Linguist

• Erica Giorda – Product Manager

AFRL Team

• Dr. David Van Veldhuizen

• Kenneth Norman

• Stacie Tawney

• Urie Reed

2. Compendia product Overview

Origin Story – NSRC Collaboration

What’s in a name?

A compendia is plural for compendium.

A compendium is a collection of concise but detailed information about a particular subject, especially in a book or other publication.

A collection of things, especially one systematically gathered.

COMPENDIA Oxford Languages

Titan’s Compendia is a flexible and modular orchestrator for assembling knowledge.

GET DO

Information Process Knowledge

• Data professionals use Compendia’s AI & ML-enabled ETL environment to ingest and transform unstructured, semi- structured, and structured content into useful information.

• Data curators use Compendia to rapidly generate high quality metadata and prepare information for use in applications like search, visual analytics, business intelligence, and records management.

• Compendia’s NLU-enabled precision search and recall capability yields more useful answers to complex research questions helping analysts and specialists to get to insight faster.

DOE National Lab ROI: 2,000 researchers = $34M wasted annually

A data enrichment & search solution that enables organizations to more effectively organize, discover, and derive insights from their information resources

Compendia for research

Key technology – Optical character recognition

Key technology – natural language Processing / Understanding

Deep Linguistic Analysis

Supports sophisticated natural language understanding by parsing each document sentence into tokens, lemmas and parts-of-speech

Knowledge Graph Empowered

Resolves ambiguous terms to their precise meaning, and determines overall document sentiment

Entity Extraction & Linking

Recognizes People, Companies, Locations and other Entities

“Fire” as the event of something burning vs. “Fire” as a dismissal vs. “Fire” as discharging ammunition vs. “Fire” as stir up or inflame

Relationships

Disambiguation

Text Metadata-Enriched Text

Knowledge Graph

NLP/NLU Engine

Language Understanding

Linguistic Projects

Text Categorization Entity Extraction

Applications & Products • Search • Content Management • Text Analytics • Natural Language API • Compendia

Studio

Development Environment

Natural Language Processing / Natural Language Understanding (NLP / NLU)

Natural Language Processing for Document Understanding

Deep Linguistic Analysis

Text Subdivision

Part of Speech Tagging

Morphological Analysis

Lemmatization

Syntactic Analysis

Semantic Analysis

Document Understanding

Topic Classification

Entity Extraction

Relationships

Sentiment

Standard Extractions

Custom Extractions

Knowledge Graph

• Natural Language Processing pipeline semantic network • English Knowledge Graph contains ≈450k concepts and millions of links • Can incorporate unique customer knowledge and focus areas

Comprehensive Off-the-Shelf Linguistic Projects

*Additional specialized taxonomies and custom entity extractions are available (including military and intelligence entities).

• General Topics • Finance • National Security & Intelligence • Crime • Emotions & Behaviors • Geography

Standard Taxonomies*

• People • Organizations • Companies • Geographic Places • Postal & Email Addresses • Phone Numbers • Social Media Mentions • Hashtags • Vehicles

Standard Entities*

Studio Development Environment

• Linguistic project builder for entity extraction and text classification

• Integrated with natural language understanding engine

• Inline knowledge graph browser with search capabilities

• Powerful editor with syntax highlighting and autocomplete with IntelliSense

• Text annotation tool for setting target entities and categories

• Quality dashboard for measuring precision and recall with respect to the targets defined in Studio

Compendia’s NLP/NLU Powers Precise Content-Level Search

Document Content Tags

Compendia’s NLP/NLU Powers Precise Content-Level Search

Taxonomy Categories

Compendia at a glance: AI-ENABLED DATA ENRICHMENT & Search

TAKE THIS TO ENABLE THISTURN IT INTO THIS HIGHLY UNSTRUCTURED DATA

RESEARCH ARCHIVE HIGHLY STRUCTURED DATA ENRICHED THROUGH

ML WORKFLOWS & NLP/NLU HIGHLY EFFICIENT, USABLE, SCALABLE DATA ARCHIVE;

DISCOVERABLE ENTITIES & RELATIONSHIPS

21 3

Current Architecture UI / END USER CLIENTS WEB PERSISTENCE

Expert Studio

AWR UI

NNCompendia UI

NNCompass UI

Compendia Dashboard UI

NiFi UI

Authentication AWR

Logging

Processor

Synch PostgreSQL

elastic

NAS

Authentication

Logging

www

Load Balancer

www

Load Balancer

JETTY//:

TOMCAT

NiFi

Logging

Data Provenance

Flow Files

Metadata normalization

Classification extractor

Custom data source API

De-duplication

Permission Persistence

Processors

Expert.nlp

elastic

3rd party CSV, xls, json, xml, Etc.…

Structured data, Metadata files

Formatted Results

Search Results

https – json doc text, awr session data,

normalized permissions

Metadata for Search

Store Doc text (Unencrypted)

Search Results

Text/ results

Get Doc text

Processing queue

Data sources and files

Documentum, Windchill, SharePoint, shared drives, PDFs, Word documents, PowerPoint, Etc.

Deployment options

Cloud native Multi-cloudhybrid On-premise

1 2 3 4

Features Pro Core Basic Ingestion, Extraction, Preparation X X X AI-enabled Metadata Extraction X X X Precision Search & Recall X X X Analytics Orchestration X X X Export X X X NLU/NLP Enrichment X X Knowledge Graph X X Linguistic Project Development X Support Pro Core Basic Titan Omni-channel Support X X X

COMPENDIA for research - LICENSING MODELS

Use or disclosure of data contained on this sheet is subject to the restriction on the title page of this proposal.

Demo

Q & A

AFRl/STO questions • Clearances • More than 20 cleared personnel

• Linguists, Engineers, Architects, Data Scientists, Executives

• Mostly TS/SCI, a few CNWDI

• Examples of Compendia running on closed systems • [Compendia] LANL, LLNL

• [NNData] AFRL, DARPA

• [ESUSA] FBI, DIA, DTRA, IC Agency

• Performance feedback & tuning • Direct access via cleared personnel

• In collaboration with cleared govt personnel

• Hybrid approach

AFRl/STO questions

• Is Compendia already listed on some existing “approved government software list” or within some system’s “authority to operate?”

• ATO for Production instance of Compendia on classified at LANL is pending • [ESUSA]

• Prod on unclassified network for FBI

• Prod on SIPR for DTRA

• Prod on JWICS for an IC Customer

• Dev on JWICS for DIA ; Pending on Production/JWICS

• [NNData]

• Prod on SIPR for USAF

• Prod on NIPR pending with Army

AFRl/STO questions • UDF…?

• The Unified Data Framework (UDF) is an extensible tuple store that allows Compendia to unify and normalize data across many different source types, including unstructured data, semi-structured and structured data. The normalized data can then exported or streamed into multiple formats and structures, including SQL, NoSQL, JSON, XML and Elastic indexes.

• Can Compendia work with (“learn”) from non-standard, non-proprietary data formats? • Yes, but this is done by training the OCR on the newly encountered format. The system doesn’t learn this

automatically.

• Describe your access framework… • Permissions Persistence Subsystem: Extracts, normalizes, synchronizes, and persists container / folder-level user

permissions.

• Designed to query each independent source system at a pre-determined interval to extract the containers / folders where each document resides, along with a list of the users who have permissions to either “read” or “browse” the associated content.

• If permissions are modified for a particular container / folder in the source system, those changes propagate to the entire Compendia system.

AFRl/STO questions • User interface via browser…?

• Yes

• How are instantiation improvements made? For example, how do feedback loops operate when customizing linguistics packages (by incorporating existing customer ontologies, taxonomies and entities) from closed systems? • Our NLP/NLU is done manually (rules-based), meaning that there isn’t any ML that is feeding back to improve the linguistic project

automatically. There are several benefits to this approach:

• It’s fully ‘explainable’, not a black box

• It doesn’t rely on high volumes of data for training. Rather, a SME can begin developing an NLP project leveraging just their knowledge and expertise without the need to annotate data and build a large training set

• We have a set of off-the-shelf (OTS) projects ready to go, projects that have been developed over multiple years based on our expertise in government, finance, banking, insurance, pharma/life sciences, and publishing.

• These linguistic projects cover six different taxonomies for document categorization (Crime, Finance, Geography, General Topics, Emotions & Behaviors, National Security & Intelligence). We offer 61 entity extractions, covering standard entities (people, places, organizations, emails, phone numbers, etc.) as well as many custom extractions (MIDB codes, academic institutions, military facilities, etc.).

• Because our technology is rule-based, these OTS projects can serve as a solid starting point for most customers, with the ability for users to easily tune and build on these projects for their specific needs.

AFRl/STO questions • Multiple copies of an existing database may exist on different systems… Consider the use case of

a large database that operates independently – BUT – that at regular intervals is “imported” into an even larger database operating at a different (e.g., ‘higher’) security level? Can the lower-level instantiation be imported and then “learn” the additional information?

• Compendia’s architecture enables rapid onboarding of new services from technology partners or customers / government entities, e.g., “Bring your own algorithm” capability.

• We can segment/compartmentalize to keep private/sensitive components separate, etc.

• Is a separate Compendia instantiation required? • Yes… Compendia is not a PL-5 / cross-boundary authorized system, however we enable MLS through the

extraction and normalization of Defense-specific classification markings into the ISM CAPCO standards.

• Are there additional costs for multiple instantiations? • Generally, yes but our licensing model is flexible…

AFRl/STO questions • What other solutions are you aware of that you believe we should evaluate to insure best

value to the government? • Palantir – Focused on dynamic analysis leveraging real time data where precision search and recall is a feature set rather than a

focus

• IBM – DIY options but requires significant expertise to correctly chain together replicable functionality covered with Compendia

• Lucidworks – very commoditized and not likely to handle any custom requirements

• What do you believe we should know, regardless of whether we asked? What questions to you believe are “the right questions” for us to be asking, not only you, but others we encounter? • Compendia’s architecture allows for the integration or reuse of independently developed AI models, i.e., support for “Bring your

own Algorithm"

• Compendia’s available off the shelf linguistic packages and entity extractions that can be extended/adapted via a straightforward rules-based approach

• Compendia’s ability to classify items based on security markings is unique in the market

Titan questions • Use cases / business problems?

• Types and volume of content, source systems?

• Nature of that content in terms of subject matter?

• Is there a need to do scanning or is your content already digital?

• Would AFRL provide all the hardware or would they want to acquire that via Titan?

• Any technical / OS level constraints? (e.g., red hat, docker, k8s, etc.)

• Do you anticipate integrating your own algorithms, projects, models, etc.?

• What would your deployment look like in terms of prod vs dev, unclass vs classified, etc., # of users, etc.

• Is an acquisition forthcoming? When?