Edwin Lisowski, Co-Founder and Chief Growth Officer of Addepto – Interview Series

Edwin Lisowski, Co-Founder and Chief Growth Officer of Addepto, oversees the company’s strategic growth, business development, and marketing. He brings extensive experience in data architecture, AI-driven strategy, and analytics consulting, combining technical expertise with a strong focus on scaling AI adoption and business transformation initiatives for global clients.

Addepto is a Warsaw-based consulting firm specializing in artificial intelligence, machine learning, data engineering, and business intelligence solutions for enterprise clients. The company helps organizations turn raw data into actionable insights through end-to-end AI strategy, proof-of-concept development, and production-ready model implementation. Working across sectors such as finance, logistics, manufacturing, and insurance, Addepto emphasizes tailored solutions and long-term partnerships to help clients harness AI for measurable business impact.

What inspired you to co-found Addepto back in 2018, and what gap in the market were you aiming to fill?

Back in 2018, we kept seeing two extremes: big vendors selling “one-size-fits-all” AI and, on the other side, internal teams stuck after a few PoCs because they lacked data engineering and MLOps muscle. We built Addepto to be the team that connects strategy → data plumbing → models → production, especially for data-heavy industries. That full-stack approach is still our DNA.

Which of Addepto’s service areas — computer vision, NLP, machine learning, or data engineering — has seen the fastest enterprise adoption, and why?

Over the last 18-24 months, NLP/GenAI has moved fastest in the enterprise (search, assistants, document processing) because it maps directly onto knowledge-work ROI and can start from foundation models. Industry surveys show a broad step-up in AI use in 2024, with GenAI-led use cases scaling across functions.

Many companies struggle to move from proof-of-concept AI to production systems. How does Addepto help them bridge that gap?

We treat production as a discipline, not a phase: discovery workshops, data contracts, reference architectures, CI/CD for models, observability, and “day-2” operations (drift, cost, guardrails). Concretely, we standardize MLOps and refactor PoCs into microservice endpoints that fit the client’s stack (Databricks/Spark, Kubernetes, existing BI). That’s how we consistently ship beyond demos.

Generative AI is now central to your offerings. How do you decide when to apply foundation models versus custom model development?

Our decision tree is pragmatic:

• Start with foundation models when time-to-value, broad language tasks, and variability dominate.
• Move to fine-tuning or adapters when domain terminology or tone precision is critical.
• Build custom models when latency/cost/IP control matter, data is proprietary/structured, or edge constraints apply.
  This mirrors where enterprises are going: fewer “experiments,” more fit-for-purpose architectures.

In 2024, you launched ContextClue as a dedicated knowledge management platform. What pain point convinced you that the time was right for a separate product?

Engineering clients kept asking the same thing: “Our CAD, PLM, ERP, and docs don’t talk, can you make them think together?” We’d solved it repeatedly in projects, so we productized the pattern. 2024 was the right moment because GenAI made retrieval and authoring usable for engineers (not just data teams). We announced and began rolling it out in that timeframe.

ContextClue integrates with CAD, ERP, PLM, and technical documents. Which of those data sources is hardest to unify, and how do you solve it?

CAD is the toughest: binary/proprietary formats, versioning, assemblies, and spatial context. We normalize CAD alongside PLM/ERP metadata, then map everything into a knowledge graph so parts, systems, specs, and procedures resolve to the same entities. That’s the backbone of ContextClue’s ingest pipeline.

The platform supports semantic search and document generation. How do you ensure accuracy and trust in those outputs for engineering teams?

Three layers:

• Grounded retrieval (schema-aware RAG over the knowledge graph) with citations to the source artifacts.
• Policy + testing (evaluation suites in CI, red-team prompts, regression tests).
• Human-in-the-loop for critical outputs (SOPs, compliance docs). We even open-sourced parts of our evaluation and graph-extraction toolchain to make this auditable.

What makes ContextClue distinct from other knowledge management tools in heavy industry and engineering ecosystems?

It’s engineering-native: it doesn’t just “search files,” it understands assemblies, dependencies, and change impact, linking CAD/PLM/ERP and maintenance history into an actionable graph. Competing KM tools often stop at indexing; ContextClue unifies structure + semantics and outputs both human-readable docs and machine-readable models (for digital twins, planning).

How do you see ContextClue evolving with the rise of multimodal AI, especially in combining text, schematics, and 3D models?

Two directions are already in motion:

• Vision-over-CAD & schematics: extracting topology, callouts, and BOM links to ground answers in drawings.
• 3D alignment: linking knowledge nodes to 3D coordinates/Omniverse views so maintenance or planning queries resolve to the right spot in the model. Expect richer agents that navigate parts, versions, and procedures across modalities.

Looking to the future, how do you see Addepto and ContextClue shaping one another’s growth, and where do you envision their combined impact on industry in the next decade?

Addepto will keep pushing the frontier, productionizing multimodal/agentic systems responsibly, while ContextClue turns that R&D into repeatable value for engineering teams. Together, we aim to cut “knowledge waste” (time lost searching/re-creating) at scale, measuring outcomes like engineering cycle time, rework rates, and audit prep time across plants and programs. The market’s moving from “many pilots” to “fewer, higher-value rollouts,” and we plan to be the partner and platform that consistently deliver those wins.

Thank you for the great interview, readers who wish to learn more should visit Addepto.