SoftServe Poland

🎟️ Register now for free [CLICK] 🎟️

📊 What’s on the agenda?
17:30 – 17:45 Come and integrate!
17:45 – 17:50 Official Start
17:50 – 18:20 Talk#1 "When "probably" beats "exactly" by Roman Dryndik, Senior Big Data Software Engineer at SoftServe
18:20 – 18:40 Let’s break the ice 🤝
18:40 – 19:10 Talk#2 "From Pixels to Pavement: Building AI Agents for Location Discovery" by Eryk Kądziela, Data Scientist at SoftServe
19:10 - 20:00 Networking and pizza 🍕

Talk #1 "When "probably" beats "exactly"
In data-intensive systems, accuracy is often optional — latency and memory are not. We will explore probabilistic data structures that enable fast, memory-efficient decisions. We’ll break down Bloom filters, HyperLogLog, and sketches, and show how embracing approximation leads to more efficient and practical data processing.

Talk #2 "From Pixels to Pavement: Building AI Agents for Location Discovery”
How can you uncover a property’s exact real-world address from just a single photograph—without any metadata? In this session, you’ll explore the emerging field of automated OSINT (Open Source Intelligence) and see how you can harness Large Language Models not just for what they “know,” but for how they think. Imagine yourself as an “AI detective.” You’ll learn how to guide an intelligent agent that examines architectural features, visible landmarks, and contextual clues—then orchestrates a sequence of specialized tools, from reverse image searches to geospatial databases, to triangulate a likely location. Through our feasibility study on New York City buildings, you’ll see how multi- modal data and iterative validation can help you achieve high-confidence geolocation. By the end, you’ll walk away with a clear understanding of this agentic framework, key takeaways from our experiments, and how you can apply these principles to push the boundaries of autonomous visual intelligence.

⚡ Save your spot [CLICK] ⚡

Register here: [CLICK]

📊 What’s on the agenda?
17:30 – 17:45 Come and grab a drink!
17:45 – 17:50 Official Start
17:50 – 18:20 Talk#1 "Snowflake Patterns: From EBCDIC Parsing to Automated Impact Analysis" by Paweł Jabłoński, Data & Analytics Technical Architect
18:20 – 18:40 Let’s break the ice 🤝
18:40 – 19:10 Talk#2 "LLM domain adaptation" by Piotr Kruczek, Data Scientist
19:10 - 20:00 Networking and pizza 🍕

Talk #1 Snowflake Patterns: From EBCDIC Parsing to Automated Impact Analysis" by Paweł Jabłoński, Data & Analytics Technical Architect
Legacy migrations face two challenges: decoding EBCDIC/COBOL data and avoiding regressions during development. In this session, you'll learn about a native Snowflake architecture that solves both.
You'll see how Vectorized Python UDFs parsed hex-encoded structures 97% faster than standard tools (cutting 24 hours to 48 minutes). You'll also learn a novel method for Automated Impact Analysis — using Zero-Copy Cloning and deep JSON comparison to validate mappings for a proprietary target system, as one more layer of regression testing.

Talk #2 "LLM domain adaptation" by Piotr Kruczek, Data Scientist
LLMs work great in general domains, but things get harder when the data is limited or highly specialized. In this talk, you'll discover practical ways to adapt models to such settings, explore experimental results, and learn what worked — and what didn’t — in practice.

⚡ Save your spot now [CLICK] ⚡