Person
1 episodes · 24 quoted claims
EPAM is a global professional services company that, according to Pavel Veller, operates in more than 50 countries and has existed about 32 years.
“EPAM is a professional services organization. We're global, we're in 50 something countries, 50 zero. People globally work with clients. We have been for, I think, 32 years to date.”
Pavel Veller says EPAM buys off-the-shelf software for non-differentiating functions like accounting and CRM, but builds software for how it operates.
“things that do not differentiate you, like an accounting software or like CRM, you would go and buy off the shelf, things that differentiate you.”
EPAM has built its own internal operations software since as early as the late 1990s.
“All of that software has always been our own since as early as late 90s, and we keep iterating on that software for ourselves.”
EPAM's DIAL platform is open source.
“by the way, it is open source. You guys can also go look, you know, download and play with it”
DIAL began as a load-balancing, model-agnostic proxy that spreads requests across model deployments with usage limits.
“From day one, dial was a load balancing model agnostic proxy, right? So every model, every deployment has limits, you know, tokens per minute, tokens per day, whatever requests per minute.”
DIAL can log all prompts in an organization so they can be analyzed to see what problems teams are trying to solve.
“We can log everything that is prompted in the organization, we can do analysis on that separately, because that's very helpful to know what kind of problems your teams are trying to solve.”
Pavel Veller says DIAL is evolving to use MCP and runs at roughly twenty-some EPAM clients.
“Now it's evolving clearly towards what MCP can bring in, cause now you can connect a lot more things to it through MCP. So I think it's running at like 20 something clients by now.”
DIAL apps, which implement a streaming API that returns text plus interactive UI elements, predate MCP.
“We have dial apps, um, they predate MCP, but the idea is that you so dial itself has a contract, um, an API that you implement.”
Pavel Veller describes MCP as a way to connect existing software to large language models, comparing its role to HTTP for browsers.
“MCP allows to connect the existing software world to LLMs in a way.”
Pavel Veller says the IntelliJ IDE exposes itself as an MCP server.
“IntelliJ exposes itself as an MCP server to do things that ID IDE can do.”
Pavel Veller says MCP-style connectivity did not exist about three months before the recording, when everyone wrote their own integrations.
“So it's this connective tissue that did not exist three months ago, three months ago. Everybody was writing their own”
EPAM has an agentic developer called CodeMie, part of its AI/RUN set of agentic systems.
“we have our own agentic developer, it's called AI run code me, uh, cause AI run has multiple different agentic systems.”
Pavel Veller said CodeMie ranked fifth on the SWE-bench Verified list at the time of recording.
“By the way, you guys can go to SWE bench and look at verified list. I believe code me as of now is, uh, takes 5th.”
The CodeMie team built an MCP server exposing its assistants so DIAL can orchestrate them as tools.
“Hey, code me, give me an MCP server for you, which is what they have done.”
Pavel Veller said Cursor had recently introduced a Max mode billed per use, which runs more iterations and spends more tokens.
“Uh, they've introduced very recently cursor on on Sonnet 3.7 and now on Gemini 2.0. I think they have a max mode, which is pay per use”
Veller's failed productivity experiment used EPAM's CodeMie agent, which iterated for about ten minutes before he scrapped its output.
“It was the code me agentic developer that we have.”
Pavel Veller works in a monorepo containing about 20 separable projects.
“I have a mono repo of like 20 different things in it that could have been separate repos of their own.”
Veller observed EPAM's developer agent sometimes failing on a large codebase by exceeding token limits.
“I actually saw our own developer agent occasionally choke because it attempts to read too much and it just chokes on. Like tokens and limits”
EPAM built a talk-to-your-data app for DIAL the previous year, using a semantic layer to turn questions into queries and charts.
“one of the apps we built as an example earlier, it was last year, uh, was like, talk to your data.”
EPAM's internal applications stream data and events into a global data hub resembling a large Kafka cluster.
“all of those applications, they stream their data and their events out into a global data hub.”
EPAM's internal API service accepts queries in RSQL, and Veller says an MCP server for it would be easy but not very useful.
“Com speaks RSQL. I can build a server that will take your query, create RSQL.”
EPAM aimed for talk-to-your-data to democratize analytics so users avoid waiting a week for a new Power BI report, but paused because it was hard to generalize.
“you can just come into dial and say, Hey, you know, show me this, this, this and this.”
Pavel Veller says a semantic layer, meaning business descriptions of tables and columns, helps an LLM write correct database queries.
“if you have descriptions of your columns, of your tables with business meaning, then connecting that semantic layer with some data samples to LLM will allow it”
Pavel Veller recommends engineers focus on fundamentals and critical system thinking to prepare for AI's future.
“I always say fundamentals and then critical system thinking.”