Your Meet-Cute Story…

Every good relationship starts with an introduction. So, meet Jasmine Agentic. She is our AI—and your expert on everything. She employs advanced agent technology to perform deep reasoning across the entirety of our new AI-native private document store. Informed by Energy Litigation Services Group’s well-honed research methods and continuous feedback, Jasmine is uniquely effective at handling litigation content.

Energy Litigation Services Group built Jasmine as our deep research assistant and AI-native document store to reduce our time spent on administration of case files and facilitate our ability to deliver critical expert insights.

Who is Jasmine Agentic?

Jasmine is the name and personality of the human interface presented as an assistant that communicates with you by old fashioned email. Here’s how she works: you pose a question, Jasmine identifies the important aspects, rips through thousands of privileged or public documents to find what matters and then synthesizes a draft answer. She redrafts that answer until she is confident in it, or else she will tell you straight up where and why the case file is weak on an aspect.

How Does Jasmine Work?

Jasmine requires no onboarding, no new apps for you to learn, or new code for us to write. Jasmine is an 'orchestrating' AI agent for your request. She employs other AI agents behind the scenes to pull documents from the web or cloud drives, ingest and classify documents and read images, search and retrieve relevant text, maintain records, and the like. Jasmine has been trained and tuned with system instructions to follow our expert research and analysis techniques. Moreover, when prompted with incomplete context, Jasmine is learning to fill in the unspoken assumptions based on continuous feedback from our experts. While we explore this new technology, our existing systems for document management all remain in place, making Jasmine another avenue to find insight.

What is an AI Agent or Agentic AI?

An AI agent is software that can pursue a goal by itself: it observes (reads data), decides (plans), and takes actions (calls tools/APIs, writes files, sends emails), then uses feedback to repeat the loop. An agent is distinct from a chat bot which just answers a prompt usually in one pass based on its inherent knowledge and/or knowledge specifically attached to the request. Legacy software vendors are busy patching code to allow AI agents to act on their systems to perform tasks. While that improves productivity, the investment can be less (and results better and cheaper) to start over with an AI-native system.

What is Deep Research?

Deep research is multi-step workflow that turns a question into a supported conclusion that became public in early 2025 by Gemini and later Chat GPT. An AI agent plans, retrieves, reads, reasons, verifies, and cites—iterating until it can justify an answer. This function is available as quota-limited option on several major AI platforms. Such services, however, cannot scan across private case files, rendering them useful primarily for analysis of public information. Given the vast amount of public information, those deep research engines are not designed to find the “needle in the haystack.”

What is an AI-Native Document Store?

It is Energy Litigation Services Group’s solution to the deep research problem for our litigation work. We started from a clean sheet unencumbered by any legacy document platform. The tech stack is components selected with the aid of AI (GPT-5) with 90% of the stitching code needed written by AI (xAI Grok Code Fast - Preview). That produced a document store that retains and manages the documents for how LLMs use them—not humans. For example, storage (regardless of source file type) is held in vector representations, tokenized, chunked embedded text (i.e., the form LLM agents search and read more efficiently). Ironically, it was quick to develop since most of the code for a traditional document store relates to the user interface to make it intuitive for humans to run complex searches or organize their documents. The AI designed ours with no human interface and only a channel that an AI agent can call. But that channel can execute any search code available in the tech stack (e.g., SQL, graphql, key word, BM25, semantic search, reranking), supporting the nondeterministic nature of AI. The result is the case file can be searched however the AI thinks best, usually multiple different ways in parallel. But yes, it still supports human search, too.