Technology

Deloitte launches Edisyl for enterprise AI context packs

Deloitte launches Edisyl for enterprise AI context packs

Key Takeaways

  • Edisyl is Deloitte’s newly created AI‑focused subsidiary, launched after acquiring Flipside Crypto’s AI business in July 2024.
  • Its flagship offering, the Knowledge Pack, merges a curated Domain Pack with a client‑specific Proprietary Pack to inject contextual expertise into any enterprise‑grade generative AI model.
  • The initial rollout targets Fortune 500 firms and Deloitte’s own practice groups, with a full‑time team of 24 former Flipside employees now operating from Boston.
  • Unlike many vendor‑locked solutions, the Knowledge Pack is designed to be plug‑and‑play across existing LLM platforms (e.g., OpenAI, Anthropic, Azure OpenAI).

Deloitte’s Strategic Move into Enterprise AI Contextualization

From Acquisition to Spin‑out

In July 2024 Deloitte completed the purchase of the artificial‑intelligence division of Flipside Crypto Inc. The deal separated Flipside’s crypto‑data arm, which was sold to Sonaverse Inc., while the AI team and its IP were re‑branded as Edisyl Inc. The new company is headquartered in Boston and operates under the leadership of former Flipside CEO Dave Balter, alongside COO Kyle Fugere, CTO Jim Myers, CFO Andrew “Mo” Septimus, Chief Data Scientist Eric Stone, and SVP of People Success Gabriela Serret‑Campos.

Why Context Matters for Enterprise LLMs

Large language models (LLMs) excel at generating human‑like text but often lack the precise, regulated knowledge required in regulated industries such as finance, health care, or supply chain management. Deloitte’s answer is to embed domain‑specific expertise and company‑specific terminology directly into the model’s prompt engineering pipeline, reducing hallucinations and improving compliance.


The Knowledge Pack Architecture

Component Purpose Typical Size Update Frequency
Domain Pack Curated, expert‑validated knowledge for a vertical (e.g., banking, pharma) 5‑15 GB of structured text, ontologies, and policy documents Quarterly review + ad‑hoc updates
Proprietary Pack Organization’s internal lexicon, SOPs, data dictionaries, and policy libraries 1‑8 GB (depends on enterprise data volume) Monthly sync with client’s knowledge‑management system
Combined Knowledge Pack Seamless overlay that can be attached to any LLM via API or embedding injection 6‑23 GB total Automated versioning; live patching for critical changes

How It Works

  1. Ingestion – Edisyl’s ingestion engine parses PDFs, CSVs, knowledge‑base APIs, and internal wikis.
  2. Normalization – Text is token‑standardized, entities are mapped to a unified ontology, and duplicate content is de‑duplicated.
  3. Packaging – The Domain Pack and Proprietary Pack are merged into a single vector store that can be called as a “contextual prompt” by any downstream LLM.
  4. Deployment – Clients integrate the Knowledge Pack via a lightweight SDK (≈200 KB) that injects the relevant vectors at inference time, requiring no model retraining.

Early Adoption & Design Partnerships

  • Deloitte’s own consulting units have piloted Knowledge Packs for risk‑assessment bots used by the firm’s audit practice.
  • Fortune 500 partners spanning banking, manufacturing, and energy are co‑creating vertical‑specific packs, with three pilots slated for production rollout in Q1 2025.
  • The 24‑person Flipside team transitioned intact, preserving deep expertise in data pipelines, prompt engineering, and LLM safety.

Comparison with Traditional Enterprise AI Solutions

Feature Edisyl Knowledge Pack Vendor‑Specific Fine‑Tuning (e.g., OpenAI Fine‑Tuned Models) Third‑Party Prompt Libraries
Model Agnosticism ✅ Works with any LLM that supports external context ❌ Tied to the provider’s model ✅ Generally provider‑agnostic, but limited depth
Update Cadence Automated monthly for Proprietary Pack; quarterly for Domain Pack Requires full re‑training (weeks) Manual, ad‑hoc
Compliance Controls Built‑in policy tagging and audit logs Dependent on provider’s tooling Variable, often missing
Implementation Overhead SDK integration < 5 hours Model training pipelines (days‑to‑weeks) Simple copy‑paste, but limited customization
Cost Structure Subscription per pack (≈$12,000 / month for Fortune 500 scale) Per‑token fine‑tuning fees (≈$0.12 / 1k tokens) Free‑to‑low cost, but higher operational risk

Outlook and Market Impact

Edisyl’s launch positions Deloitte to capture a growing slice of the enterprise‑AI contextualization market, projected to exceed $4 billion by 2028 according to IDC. By offering a plug‑and‑play Knowledge Pack, Deloitte sidesteps the heavy compute costs of model fine‑tuning while delivering the compliance and accuracy demanded by regulated sectors.


Bottom Line

Deloitte’s spin‑out, Edisyl, transforms the way large organizations harness generative AI by layering vetted domain expertise and proprietary corporate knowledge onto any LLM. The modular Knowledge Pack reduces hallucinations, accelerates time‑to‑value, and maintains strict governance—key differentiators in an industry still wrestling with AI reliability. With a seasoned 24‑person team, early Fortune 500 pilots, and a Boston‑based headquarters, Edisyl is poised to become a cornerstone of Deloitte’s AI practice and a credible alternative to costly model‑specific fine‑tuning.

For more details, visit deloitte.com.

Related Articles