Here we will share the Innovate UK project reports on the development progress.
Q1 Report 04 2023 to 06 2023
Overall Progress and Key Achievements
The project’s key focus for this period was to build the functioning prototype and identify potential users who could validate the use case and the product. We have defined end-to-end user flow for the CV use case to find the right people in engineering projects, built an AI engine for enhanced contexted search and role-specific queries, and organized a demonstration to INCOSE EMEA on 28 June (WEBINAR: Personalized Semantic Search using the Digital Systems Engineering), after which we collected the feedback. The overall perception of the first potential users in (AEIS, the Spanish Chapter of INCOSE) is positive; they show interest and consider using it, which means we can validate a feature map and requirements specifications with them.
We further clarified and externally checked the business model with the prototype solution mentioned above for the donations-driven roadmap and corporate subscriptions. The initial segment for the market is GitHub and Logseq user communities; the revised market segment value assessment shows the financial projections’ viability and attendance.
The overall system architecture and use case are finally stabilized, thus enabling the development of the system requirements specifications and feature map, which will be completed in the coming weeks without significant risks of major changes and rework downstream. The solid case for Atomic Server usage has been built and partially implemented. Aligned use case, architecture, and Atomic Server usage allowed us to baseline the project plan, which will be essential to manage the risk profile for the rest of the project.
The prototype that processes role-specific queries is operating, and we publish the code in the GitHub repository using freshly built infrastructure CI, continuous deployment pipeline, and infrastructure security mechanisms. The accomplishment of this task enables connecting outside users to the system and collecting their feedback, which is untainted with nonfunctioning or broken deployment.
As was planned for this period, we:
- Finished project management plans development and baselining.
- Started developing a preliminary feature map and requirements specification for the CV generation use case.
- Developed and tested the prototype for the CV use case and an AI engine for enhanced contexted search and role-specific queries.
- Built infrastructure CI, continuous deployment pipeline, and security of the infrastructure.
- Started creating the go-to-market strategy.
- Started developing a specification for the validated requirements for Atomic Server and agreed on the most critical of them with the Atomic Server team.
The team communicates and works well, holds to its commitments, uses agreed tools and templates, addresses risks, and actively participates in meetings.
Technical Progress
Work Package Number 1: Project Management
Work Package Objectives:
- Ensure the smooth and effective implementation of the Eurostars project.
- Successfully met all the project’s R&D goals and objectives within the budget and scheduled timescales based on SMART evaluation criteria.
Description of work this period - Collect and upkeep the project case – project plans-to-date and reports, financial forecasts and timesheets, risk registry, etc.
- Plan, control, and report on deliverables and milestones and provide objective evidence of the project’s progress.
Progress towards the Deliverables for this Work Package
All relevant information has been collected and archived. Final versions of documentation have been submitted to Innovate UK.
All required documents have been uploaded, and the onboarding procedure has been completed. Timesheets are prepared and verified against the schedule and budget. Lessons learned about the specifics of operating the Innovate UK project are extracted and will be implemented as part of standard operating procedures in the subsequent periods.
Description of planned activity for next quarter.
Controlling the performance baseline, reporting, and proactive risk management, especially marketing, technical, and systems integrations. Stabilize document and information management flows and archiving policies.
Work Package Number 2.1
Definition of product requirements and specifications
Work Package Objectives:
-
Develop system definitions (requirements, architectures, and product planning and marketing) for the Atomic Server prototype with Terraphim search capability detailed enough to plan and build integrated prototype TRL6 and define and validate its use cases. This task will be performed from the Terraphim development team’s viewpoint, limiting the scope for requirements and specifications for Atomic Server on AS system and user interfaces.
-
Develop approaches and plans to verify and validate the prototype and Terraphim as a subsystem.
Description of work this period -
Develop preliminary feature map and requirements specification for the CV generation use-case.
-
Develop and test prototype for the CV use-case.
-
Develop the go-to-market strategy - revision 1.
-
Perform prototype validation and analyse the results for the CV use-case.
-
Develop specification for the validated requirements for Atomic Server revision 1 and agree with the Atomic Server team.
Progress towards the Deliverables for this Work Package
Deliverable 3. Preliminary feature map and requirements specification. Due date: 30 July 2023
We developed several candidate use cases: The T-competence read & annotate, GitHub project taxonomy crawler and enforcer, and end-to-end user flow for finding suitable systems engineering skills using the systems engineering digital process model. Then made external evaluations of the attractiveness of those use cases and selected the latter end-to-end user flow to implement the prototype of an AI engine for enhanced contexted search and role-specific queries.
Deliverable 4. Early prototype and test results. Due date: 30 June 2023
An early prototype has been built, and after evaluating its performance against the competition that emerged in the last couple of months, we decided to make a much more functional Terraphim search engine much earlier than we planned in the project proposal. That was required because the risks associated with users validating product specifications and requirements with an early prototype lacking some key planned functionalities were unacceptably high. For example, our first demonstrations needed to have search relevance managed through a personal knowledge graph feature available. As works to deliver early prototype and production-level solution are hard to distinguish, technical implementation details are placed under the Work Package 2.4: Development of the AI search engine “Terraphim Search” section.
Deliverable 5. GTM strategy rev.01. Due date: 30 July 2023
We have prepared two drafts of the go-to-market strategy for the first two use cases. Though the end-to-end user flow we selected for the implementation and user validation was demonstrated on 28 June, the GTM strategy rev.01 should be subsequently adapted and specified during the next project period. We needed to postpone the finalization of the GTM strategy because the industry’s competitive landscape has been changing a lot during 2023, and only now has it stabilized enough for us to respond to the changes and reposition the product.
Deliverable 6. Validated requirements for Atomic Server. Due date: 30 July 2023
We formulated the critical requirements for Atomic Server. Still, until we confirmed the end-to-end user flow, we considered them unstable and technically dependent on the future implementation of Terraphim search. So now the Atomic Data team is performing the feasibility study. Once we get the confirmation on the next project meeting, scheduled on 03 July, we can develop and validate the final requirements specification.
Implemented technical enablers: Build infrastructure continuous integration (CI), continuous deployment pipeline, and infrastructure security.
Together with the Atomic Server team, some critical requirements for Atomic Server have been validated and implemented in the latest release Release v0.34.5 · atomicdata-dev/atomic-server · GitHub :
- Add a query endpoint that allows performing collection queries via an endpoint instead of repurposing the collections collection.
- Add support for Bearer token authentication; find in /app/token #632.
Infrastructure continuous integration:
1. Cost effective private cloud deployment, caddy + caddy auth, bash CGI, firecracker VM + Redis (Enterprise). Allows to create a new VM for atomic server and Terraphim. The atomic server is under 2 seconds per VM, and Terraphim AI is 6 seconds (+10 seconds sleep for a cluster to bootstrap). Terraphim minimal VM is 16 GB RAM, 32 GB disk + 16 cores.
2. CI pipeline for Terraphim platform GitHub - terraphim/terraphim-cloud-dependencies: Dependencies to run Terraphim in the cloud (or RPi)
3. CI pipeline for GitHub - terraphim/terraphim-cloud-fastapi: FastAPI python and Redis OM backend for Terraphim Cloud
4. CI pipelines for GitHub - terraphim/terraphim-ui-svelte-ts: Terraphim UI
5. CI pipeline for GitHub - terraphim/terraphim-platform-pipeline: Cloud (python) based pipeline for Terraphim AI assistant
6. CI pipeline for GitHub - terraphim/terraphim-logseq-md-parser-simple: Simple parser for Logseq to turn methodologically tagged files into knowledge graph (Rust parser)
7. CI pipeline for GitHub - terraphim/terraphim-project: Product and project management for terraphim - uses Logseq md parser simple to parse Logseq knowledge graph and publish Logseq KG into web and uses aws_op_earthly to write Terraphim “index” into S3.
Infrastructure security:
1. 1password CI operator, allows to fetch credentials from 1password connect server secure vault using OP_CONNECT_HOST and OP_CONNECT_TOKEN token. GitHub - applied-knowledge-systems/aws_op_earthly: Demo of pipeline using 1password connect cli to fetch AWS secrets
Why important?
1. Dependency for WP2.1.
2. Allows sharing infrastructure code, including environment files.
3. Makes security simpler (manageable): only one token per environment needs to be protected - i.e., production/staging/testing
4. The exact mechanism can be leveraged to store the user’s tokens and only reference them in atomic config via op://terraphim/username.
2. Set 1password connect server in Cardiff Oracle cloud https://central.terraphim.io (deployment scripts are using docker-compose + one password cli + Cloudflare edge tunnel). Dependency for above.
Summarise any variations from the Second Level Plan, giving reasons and action to recover the situation if necessary.
The plan’s current version was produced in June and accounts for proposed changes to the original schedule approved in the project proposal. Deliverable 3. Preliminary feature map and requirements specification; Deliverable 5. GTM strategy rev.01; and Deliverable 6. Validated requirements for Atomic Server will be completed in July and not in June as initially planned.
The reason behind the changes is described above, and it is significant changes in the competition and technology landscape in the AI industry affected the viability of the original plan.
Description of planned activity for next quarter
Completing Deliverable 3. Preliminary feature map and requirements specification; Deliverable 5. GTM strategy rev.01; and Deliverable 6. Validated requirements for Atomic Server.
Drafting Deliverable 7. Terraphim specification and preparation of the approach and detailed plan for Work Package: 2.5 System integration and optimization.
Work Package Number : 2.4
Development of the AI search engine “Terraphim Search”
Work Package Objectives:
- Build an advanced prototype of the Terraphim search engine as the first Atomic Server application.
Description of work this period
- Build knowledge graph-based search core prototype.
- Build an AI engine for enhanced contexted search and role-specific queries.
Progress towards the Deliverables for this Work Package
Deliverable 10. Functional prototype of AI search engine. Due date: 29 June 2023
The end-to-end prototype for Terraphim AI in a cloud was developed and open-sourced:
1. uplift Python FastAPI
2. uplift Terraphim Us svelte
3. migration of Python-based pipeline from The Pattern into the Terraphim Platform Pipeline
4. Terraphim Automata Parser in Rust
5. Bindings for Terraphim automata for node js
6. Terraphim Rust pipeline - sentence parsing in Rust
7. Terraphim Poem API - Rust replacement for Python-based FastAPI
8. Research on KV (OpenDal)
9. Logseq parsers:
1. parse export from Logseq (JSON) using Rust (private repository)
2. parse markdown from Logseq into OpenDal dashmap operator with config load (unpublished rust code)
3. Parse markdown from Logseq into CSV (terraphim-logseq-md-simple)
10. Terraphim UI:
1. updated to accommodate project managers’ flow
2. Created workers - which will allow fetching JSON, JSON-AD, Wikimedia, and Atomic Data config
Summarise any variations from the Second Level Plan, giving reasons and action to recover the situation if necessary.
We started developing a functional search prototype earlier to benchmark existing solutions. Without it, validating requirements and product specifications in this market is impossible.
Description of planned activity for next quarter
Further development of knowledge graph-based search core prototype and technical enablers:
- Rewrite python-based pipeline into Rust crates:
- sentence splitter + aho corasick (matcher)
- Plug into Poem
- wrap (a) into Node-JS bindings
- create a tokio worker to keep a queue to write into RedisGraph (step 3 in the pipeline)
- Create Tauri build (with Redis-Stack dependency)
- Integrate configuration management from Atomic
- Fetch Atomic Server config via URL and svelte code and update local (Redis) Config
- Stretch goal: Remove dependency on RedisGraph - write your own “star decomposition” and “clique decomposition” structures using the OpenDal operator
- Create Terraphim extension (patch Logseq co-pilot)
- Infrastructure: create a method to bootstrap private cloud from barebones
- Infrastructure: update VM to include zerotier network (VMs can be moved around).
High-level activities from the project plan:
- Build user knowledge graph work experience builder/enricher prototype.
- Develop user needs validation plan, collect and analyse the results for the knowledge graph-based core prototype.
- Develop and deploy all key technical enablers and dependencies required for deliverable 11, “AI search engine,” and deliverable 9, “Haystack search prototype with connectors.”
Q2 report 07/2023 to 09/2023
Overall Progress and Key Achievements
The project’s previous focus was to build the functioning prototype and identify potential users who could validate the use case and the product. The results were primarily informal, and making proper project deliverables (product roadmap, go-to-market plan, and requirements specification) from them was the first task we completed. We built and launched the website with the demo video to validate the product concept further.
Then, we prioritized different use cases from the developed roadmap and implemented end-to-end use cases with the highest priority. We performed a soft launch of the early version of the role-based search engine. We found early adopters of technology eager to test it for different scenarios – software developer jobseeker, academic researcher, and systems operator being top of them. We also started to promote the solution for corporate customers (a cyber security product development company and an NHS Pharmacy Service). As part of this job, we developed brand positioning and the standard pitch, which we tested during the presale with these corporate customers.
We also advanced with integrating Terraphim with Atomic Server into one coherent and valuable solution. However, we identified a few significant risks for the subscription components of the original business model described in the project proposal due to changes in competition. The best part of the business model, with consultancy and subscription-driven roadmap, still holds, though.
As we identified the initial segment for the market earlier to be GitHub and Logseq user communities, we started to gather early adopters from the Logseq forums to discuss and co-develop future features with them.
We continued the implementation of system interfaces between Terraphim and Atomic Server, further specifying the overall concept of operations and creating content for Terraphim specification.
As was planned for this period, we:
- Updated project management plans, especially risk management, and confirmed their alignment with technical development plans.
- Completed a go-to-market strategy, preliminary feature map, and requirements specification for the validated requirements for Atomic Server for the CV generation use case.
- Started the execution and testing of the go-to-market strategy.
- Continued the integration efforts with the Atomic Server team.
- Ahead of schedule, we started the implementation of the knowledge graph-based search core prototype and developed and tested an early version of the haystack search prototype and AI engine for enhanced contexted search and role-specific queries.
- Started preparing for dissemination and communication activities report.
The team communicates and works well, holds to its commitments, uses agreed tools and templates, addresses risks, and actively participates in meetings.
Exceptions
We are in for about 50% in the development of Work Package Number 2.4 “Development of the AI search engine “Terraphim Search,” which is significantly ahead of the planned schedule, as the planned completion date is February 2024. The reasons for this are explained below.
Terraphim Private cloud was based on the same Python codebase as “The Pattern” - winner of the build on Redis hackathon, but Redis decided not to support dependencies:
- Python-based Redis Gears were not been supported since 2022, and re-packaging old releases became a maintenance burden.
- Redis recently announced the End of Life for RedisGraph.
The latter is a core component for the Terraphim Knowledge graph, and this forced us to expedite the re-write of Terraphim Core to remove complex dependencies on Redis components and replace them with:
- Our state-of-the-art graph embeddings not only fast but retain relevant positions of the words when processing text.
- Removing Redis from embeddings allows the packaging of Terraphim AI for a wider audience into Chrome Extension and desktop clients.
- Open the Data Access Layer library to save snapshots in Redis/Local storage (Sled) or AWS S3, making dependency on Redis optional.
- Open DAL Access layer allows us to remove maintenance burden and create enablers for commercial offerings. For example, we can create a commercial service/paid feature, “snapshot Terraphim AI core into AWS S3,” for sponsors.
Technical Progress
Work Package Number 1: Project Management
Work Package Objectives:
• Ensure the smooth and effective implementation of the Eurostars project.
• Successfully meet all the project’s R&D goals and objectives within the budget and scheduled timescales based on SMART evaluation criteria
Progress towards the Deliverables for this Work Package
All relevant information has been collected and archived. Final versions of documentation have been submitted to Innovate UK.
All required documents have been uploaded, and the onboarding procedure has been completed. Timesheets are prepared and verified against the schedule and budget. Lessons learned about the specifics of operating the Innovate UK project are extracted and will be implemented as part of standard operating procedures in the subsequent periods.
Work Package Number 2.1
Definition of product requirements and specifications
Work Package Objectives:
- Develop system definitions (requirements, architectures, and product planning and marketing) for the Atomic Server prototype with Terraphim search capability detailed enough to plan and build integrated prototype TRL6 and define and validate its use cases. This task will be performed from the Terraphim development team’s viewpoint, limiting the scope for requirements and specifications for Atomic Server on AS system and user interfaces.
Develop approaches and plans to verify and validate the prototype and Terraphim as a subsystem.
Description of work this period.
- Develop a preliminary feature map and requirements specification for the CV generation use case.
- Develop and test prototype for the CV use case.
- Develop the go-to-market strategy - revision 1.
- Perform prototype validation and analyse the results for the CV use case.
Develop specification for the validated requirements for Atomic Server revision 1 and agree with the team.
Progress towards the Deliverables for this Work Package
Deliverable 3. Preliminary feature map and requirements specification. Due date: 30 July 2023
From the developed candidate use cases: The T-competence read & annotate, GitHub project taxonomy crawler and enforcer, and end-to-end user flow for finding suitable systems engineering skills using the systems engineering digital process model, we selected the last one for further implementation. We specified this use case for the first implementation that helps users prepare for the certified systems engineer exam and build their CV and development track aligned with the systems engineering handbook v.4.
Deliverable 4. Early prototype and test results. Due date: 30 June 2023 – completed before
Deliverable 5. GTM strategy rev.01. Due date: 30 July 2023
We completed the preparation and started implementing the go-to-market strategy for the first use case (INCOSE CSEP exam preparation). The first results from the performance will follow during October, and we will make further specifications and corrections based on the feedback from the users and corporate customers.
Deliverable 6. Validated requirements for Atomic Server. Due date: 30 July 2023
We agreed on the critical requirements for Atomic Server and confirmed them in the end-to-end user flow. We have a working POC integration between the Terraphim AI front end and the Atomic Server. We have a plan and steps to build a joined Terraphim AI and Atomic Server Chrome browser extension. The configured Atomic Server will drive Terraphim AI configuration and allow users to save pages from the browser into Atomic or Terraphim.
Deliverable 7. Terraphim specification. Due date: 30 Oct 2023.
We prepared the content for an early version of the Terraphim specification for internal discussion within the team on the subject of system integration. The specification will be published on terraphim.ai and https://terraphim.discourse.group/ for community engagement.
Work Package Number: 2.4
Development of the AI search engine “Terraphim Search.”
Work Package Objectives:
-
Build an advanced prototype of the Terraphim search engine as the first Atomic Server application.
Description of work this period. -
Build knowledge graph-based search core prototype.
-
Build an AI engine for enhanced contexted search and role-specific queries.
Progress towards the Deliverables for this Work Package
Deliverable 9. Haystack search prototype with connectors. Due date: 29 Dec 2023
We built the first version of the haystack search prototype that uses the system operator knowledge graph and web browser Terraphim plugin to perform concept embedding with further ingestion of the webpage into Terraphim Cloud and unifying them into a searchable haystack. We are going to test this functionality in our end-to-end user flow.
-
Rewrote Python-based pipeline into Rust crates:
-
Completed: Sentence splitter + aho corasick (matcher)
-
Completed POC: Created Rust based Open API (Poem) for Terraphim to deploy in cloud. Planned for next quarter: Rewrite in line of removing Redis, RedisGears and RedisGraph from hard to optional dependencies.
-
Completed: Wrapped Terraphim matcher into Node-JS bindings. In addition, we created a wasm binding which allows the creation of Chrome Browser extension, expanding the reach for potential dependency.
-
Majority of work done to make Redis optional dependency, now we need to re-package our releases and update continuous integration/continuous deployments pipeline.
-
Integrated configuration management from Atomic
-
Completed prototype to integrate with Atomic Server: We demonstrated we can add Atomic Server, fetched Atomic Server config via URL and svelte code and updated local config Terraphim’s config
-
Stretch goal completed: Remove dependency on RedisGraph - write your own “star decomposition” and “clique decomposition” structures using the OpenDal operator
-
Created Terraphim AI Chrome extension POC
-
Infrastructure: created a method to bootstrap private cloud from barebones
- we have a complete Python-based end-to-end deployment for the cloud, but since abandoning Python, we need to re-create private cloud
The following steps are planned and scheduled:
-
to create an automatic deployment for Rust-based Terraphim cloud (Instead of Python based)
-
to create a deployment for small devices like Raspberry PI
-
Create package and deployment for Desktop (Tauri)
-
Create package and deployment for Chrome Browser extensions
-
Create a set of deployment scripts and processes that allow the scale of the private cloud up and down
-
Infrastructure: updated VM to include Zerotier network (VMs can be moved around to different servers, improving redundancy and including failover, this feature is only required for the paid service users).
-
Infrastructure: we encountered difficulties relying on external services – like GitHub or Redis repositories. We will need to take more of the ownership of dependencies for deployment:
Completed:
- 1Password connect server set, covering security aspects of CI/CD pipelines and deployment.
- CI/CD pipeline was built with Earthly and Earthly CI, Earthly CI sunset by 1st October introduced re-work on migrating repositories CI from Earthly CI to the new Earthly Satellite product
- We built a collaboration relationship with the Earthly team and agreed on the following:
- They will help promote Terraphim AI private cloud via their blog – a technical article on how to build Terraphim Private cloud and deploy the 1password connect server.
- They will help promote Terraphim AI via their Twitter (100K subscribers) when we are ready to announce it.
Deliverable 10. Functional prototype of AI search engine. Due date: 29 June 2023 – completed during the previous period.
Description of planned activity for next quarter
Further development of knowledge graph-based search core prototype and technical enablers:
High-level activities from the project plan:
-
Build user knowledge graph work experience builder/enricher prototype.
-
Develop user needs validation plan, collect and analyse the results for the knowledge graph-based core prototype.
-
Develop and deploy the other technical enablers and dependencies required for deliverable 11, “AI search engine,” and deliverable 9, “Haystack search prototype with connectors.”
Work Package Number: 4.1
Knowledge dissemination and external communication activities
Work Package Objectives:
-
Disseminate, communicate, and exploit the project’s results.
Description of work this period. -
Disseminate the R&D and validation project results (proof of concept).
-
Communicate activities via online and social media channels.
-
Expand on joint and individual business opportunities for further scale-up upon the project’s results and deliverables.
Progress towards the Deliverables for this Work Package
Deliverable 20. Dissemination and communication activities reports. Due date 28 Jul 2023.
Continuing to communicate with INCOSE about implementing the end-to-end CSEP exam preparation and the systems engineering process model conformance checks with Terraphim Search Engine. We succeeded with the AI in the systems engineering working group and INCOSE LatAm chapter, as we agreed to demonstrate the solution in October once we fully implemented the use case.
Medium business aspect risk discovered because of the feedback from High-Value Manufacturing Catapult Skills Forecasting Initiative product progress report shared with us by Paul Shakspeare, Consultant Education and Skills on 5th of September 2023:
“We are well into developing a robust data structure using process and intelligent tools to connect a wide variety of reference data sets to support our work to understand future skills needs related to emerging innovation and technologies. We will be bringing up a web presence in the next month or so, which will provide further information about this.
The issues raised by your presentation have mostly been covered in our development process. Still, I would like to see any similar examples you have worked on and can demonstrate.”
This risk affects the central development proposal as we implemented it in the first project period. To address this risk, we switched our development primarily to personal knowledge and skills management and knowledge activation, as those aspects would unlikely be covered by corporate products HVMC and other competitors are developing.
Deliverable 21. Communication content. Due date 28 Jul 2023.
The most important part of the communication content is a new Terraphim release and INCOSE system operator knowledge graph in the project’s GitHub repository. The second part of the content is the website and demos we prepared. The third part is corporate presentations and pitches, but as they are under NDA, we cannot share the links in the report. The final fourth part relates to brand positioning results that will direct our further actions related to the pitches and sales to corporate customers.
Description of planned activity for next quarter
Perform several language-market fit experiments for further validation of the product concept. We are focusing on personal knowledge activation and knowledge management for the next period. Continue pitching to corporate customers and generating leads, testing brand positioning and pricing. Re-write the website for improved UI and UX for other user segments, except the initial engineers and developers, and start the publications on the website and all other communication channels. Start implementing the user research program—record further demonstrations of Terraphim according to the GTM plan.
Executable knowledge systems: the next best action recommender
Abstract
Existing meeting transcription and memo generation product features (like Zoom) or note-taking services (like Otter, Fireflies, tl;dr or Avoma) can produce a complete list of actions and commitments from the engineering team meeting transcript. However, those memos that promised what are not to-do lists are lists of intents. They often require additional interpretation steps on what exactly must be done to complete the action item. Such additional interpretations are necessary more often when customers or sales teams are among the participants and if there are contractors on the project. Inaccurate or incomplete translation from intents to to-do lists strongly skews effort estimates, i.e., an engineer misunderstands the amount of work when reading just a Zoom-generated action item. We call it the “trans-domain translation problem.”
How do we formalize the trans-domain translation problem? We have statements or questions in the customer’s domain (the original meeting transcript) and answers and clarification questions in the engineering domain (engineering knowledge). When we read the Zoom transcript, how do we correctly translate the list of intents and promises into the complete and proper to-do list and assign the right people to perform those tasks? Intent is not an action, and mistakes are possible. The primary factor that affects the correctness of trans-domain translation is context explication. The more context we know about each intent item, the more correct the translation can be.
Translation between domains can use three mechanisms to describe the trans-domain translation contexts:
• rules and discipline (domain logic, enterprise and industry standards, legislation);
• routines (everyday practices);
• explanations and governance (legitimation narrative).
The selected context defines what the best following action would be. We call this approach “the domain lens and the analytical lens.” First, we consider the initial text through the domain lens, and then we match identified terms and concepts with the target domain through the analytical lens. In this paper, we show the airside operations domain as the domain lens and project management and system operation as analytical lenses. In this work, we show how we model these three types of contexts, implement the trans-domain translations for actual engineering tasks, and build recommendations for the next best action depending on the selected context.
Keywords: trans-domain objects, the V-model, systems engineering and program management integration, Terraphim, Atomic Server
Q6 report
Overall summary of the project
When we started working on Terraphim AI assistant, we were among the few teams working on AI assistants in a niche market. With the public release of ChatGPT by OpenAI, the AI assistants market has exploded, and according to VC reports, 70 billion dollars was invested into AI assistants last year. This presented a challenge to marketing, expansion and commercialisation. In such an environment, one needs at least a double-digit budget in the millions to make a noticeable dent in the noise-driven hype around AI assistants. We persevered and found a niche of AI-enabling tools for people working in complex environments, such as systems engineers, project managers and specialised engineers.
During the last few weeks from the middle of September 2024, with recent reports demonstrating that Large language models-based AI assistants were not able to deliver on the promise, we see it as a window of opportunity for Terraphim to gain some traction without a substantial marketing budget, leveraging what we have developed as part of Innovate UK supported project:
• We demonstrated that even by applying Terraphim AI methodology and using domain modelling within Atomic Server, we could achieve better quality output from open-source large language models like LLama 8B in complex requirements engineering flows for the aerospace industry.
• We created a set of technical components which allow Terraphim to be integrated into existing user flows:
• Visual Studio code extension to provide additional context from Atomic Server (Knowledge Graph) into LLM
• Visual Studio code extension to provide additional context and suggestions from Atomic Server (Knowledge Graph) as autocomplete suggestions
• Visual Studio code extension to parse text and replace known concepts by linking them to a knowledge graph (explainability of the terms and language alignment)
• Terraphim unique graph-based embeddings to support user-controlled ranking and re-ranking
• Terraphim AI desktop application (Rust/Tauri)
• Allows for the users to leverage their existing personal knowledge management graphs – from LogSeq/Obsidian
• Search over different collections of documents – local or remote- while maintaining the context of the role.
• Terraphim AI private cloud instance – with REST API server allows integration into web-based pipelines and tools.
• Terraphim AI private cloud – terraphim.io, which allows hosting of dedicated instances without sharing data and maintaining data/security boundaries
• Continuous deployment of the above using Earthly
• Terraphim AI technology is open source, and Terraphim is a registered trademark in the UK/EU(WIPO) and the US.
The Atomic Server team reached the stage where they perfected developer experience, specifically for Typescript. We will help the team advertise Atomic Server wider and use it as part of our technology stack.
We will continue to pursue the development of Terraphim AI via two companies: Applied Knowledge Systems Ltd for personal use and Zestic AI for business use.
End-to-end user flow. Terraphim successfully transcribes from one domain to another:
• The end-user build is published, together with installation instructions.
• This build allows users to go from meeting notes and transcripts produced by AI tools (Zoom, Meet, Read.ai, etc.) to domain-specific checklists to help implement the action items generated by such tools.
For example, Zoom gives you an action item: “Alexander to prepare and send a report summarizing the contracted scope, completed items, and outstanding items by Monday.”
o Terraphim enriches it with links to proper documents and generates non-obvious questions to validate the definition of done and reduce the friction when delivering on commitments.
o Atomic uses those links and questions as a context for retrieval augmented generation for LLMs to generate such checklists, help answer them, and generate proposals to collaborate using different process models and SFIA framework.
• This end-to-end user flow implementation curates the Terraphim methodology we documented on the Discourse platform – describe the business and project domains, choose the analytical lens and plan the project.
Alpha release for Terraphim, integrated with Atomic Server, is packed into Visual Studio Code to provide a smooth user experience for all steps in user flow:
• Visual Studio Code LLM context provider extension for Terraphim.
• Visual Studio Code context operator integration with Atomic Server.
• Validation of user flow for different LLMs and confirmation that it works and is practical using the small ones like Llama 3-8B, which can be run locally on the notebook, which addresses the privacy aspects of our initial proposal.
Advancements in Terraphim Code:
• The code is fully documented now and is ready for further contribution by the community.
Advancements in Atomic Server:
• Further improvements to the ontology editor feature.
• Improvements in API, authentication and integrations.
Commercial exploitation:
• We have signed off on the next-stage delivery for Charm Impact, which uses Atomic Server and Terraphim as part of the proposed solution.
• We are using the implemented user flow to pitch potential clients as proof of our competencies in AI, especially trustworthy and deterministic AI. It is highly appreciated by engineering companies like geo-engineering or oil businesses.
• We continued the lead generation and selling activities. We submitted a paper for the Rust Nation conference and continue to extend our reach and network of contacts in other ways. Participation is to be confirmed.
• We submitted an abstract to Haystack EU 2024, but our talk about Terraphim AI was rejected
Dissemination activities:
• We recorded and published the end-to-end user flows for Terraphim and Atomic applications, which we will promote through our network.
• We have published an article and written a white paper on the Terraphim application as a knowledge system.
• We have entirely documented the code and published guidelines for the community on contributing to it.
As was planned for this period, we:
- Finalised project management plans and have written the final report.
- Completed the integration and validation of the end-user solution with the Visual Studio Code with the Continue plugin as an end-point user interface. We have chosen VSC because of its popularity in the software development community.
- The updated version of the roadmap has been published.
- Prepared the business case for further development of the product.
- Continued commercialisation activities – searching for potential clients, preparing the demos and RFQs, and refining the business model and pricing.
The team continues to work together after the project is completed, as we have a commercial delivery now. We further confirmed with our clients that the viable market segment for our technology stack is deterministic AI for Terraphim and perfect engineering publishing experience for Atomic, and we are progressing well in that direction.