UiPath Agentic AI ile Akıllı Otomasyonun Yeni Çağı
- 1. UiPath Agentic AI ile Akıllı Otomasyonun Yeni Çağı Agentic AI ile geliştirilen UiPath Agent mimarisini tüm bileşenleriyle keşfedin. Bağlam yönetimi, kullanıcı etkileşimi ve Orchestrator entegrasyonu gibi başlıklar canlı demolarla aktarılacaktır.
- 2. 2 Istanbul UiPath Community Chapter
- 3. 3 Etkinlik, @ChapterIstanbul tarafından sağlanmıştır. Levent Kurt Konuşmacı UiPath MVP 2024-2025 Intelligent Automation Team Lead @NTT Business Solutions [email protected] https://www.linkedin.com/in/leventtkurtt/
- 4. 4 İş hayatında gerçek bir projeye dayalı olarak, pratik ve uygulanabilir kullanım senaryosu çözümleri ve yaklaşımlarıyla dolu bir 90 dakika sizleri bekliyor. Etkinlik sonunda tüm proje sizinle paylaşılacaktır ☺ UiPath MVP'si ve İstanbul UiPath Topluluğu Bölüm Lideri Levent Kurt ile tanışın(2dk) Projeye Genel Bakış (5dk) Agent Yapısının Temelleri ve Bileşenler (10dk) Gerçek dünya kullanım senaryolarının canlı demo (60dk) • Agent Structure with all components • Context Grounding • Escalation with UiPath Apps Structure • Agent Evalualtion Sets • Publish Agents To Orchestrator • Run Agent From UiPath Studio as activity (Input and output variable is important) • Call Agents and RPAs from Maestro İyi eğlenceler! Geri bildirimleriniz memnuniyetle karşılanır.(10dk) Agenda
- 5. 6 Agent versus Workflow Agents Goals-based, act independently, dynamic decisions Robots Rules-based, act predictably, deterministic decisions
- 6. 7 What are the components of an agent? Natural Language Prompt Instructions or plan for the agent –its role, goal, and constraints; *Input/Output Arguments *System prompt with role, goals, constraints, #Error and escalation *User Prompt *Examples Context(Reference) Short and Long-term memory used to inform the plan. Context Grounding to support: • Querying from knowledgebases to ground prompts • Agent Memory so the agent can learn based on its runtime Tools (Function) UiPath tools that are invoked based on the prompt Tools an agent could use: • Activities • Automations • Other Agents Escalations Involve a human only when necessary – to help gather additional information or review input/output arguments Agent Escalation paths: • Action Center Action Apps • Communication channels (e.g. Slack) Agent Evaluation and Calibration – Designtime and Runtime AI Trust Layer with LLM Ops – Trust, Transparency, and Governance
- 7. 8 Bad And Good Use Cases For Agents Drafting and summarizing content (emails, messages) Multi-system research and data gathering First-pass customer interactions or ticket triage Orchestrating narrow AI agents into larger workflows GOOD USE CASES FOR AGENTS BAD USE CASES FOR AGENTS High-risk financial transactions Regulatory workflows with zero error tolerance Complex multi-step data processing without deterministic safeguards
- 8. 9 Well structured Agent Prompt? A high-performing agent requires instructions that clearly determine a plan for action, incorporate inputs in a well-structured way, and gives guidance on when to run tools, access enterprise context, or escalate to a human. You achieve this by writing prompts and defining agent arguments. UiPath Product Manager Template From MVP Summit #Role #Goal #Instructions #Constraints #Error and escalation
- 9. 10 System and User Prompt? System prompts System prompts allow you to describe in natural language the role, goal, and constraints of an agent. You specify any rules for it to follow, and add information about when it can use certain tools, escalations, or context. The system prompt helps the agent form a plan that it uses and adapts over time from interactions with tools, robots, and humans. A good system prompt suggests a sequence of steps, addresses certain cases, and tells an agent when it should call tools or raise escalations. User prompts User prompts allow you to structure how the inputs and arguments are passed to the agent. You can also show in the user prompt how you refer to certain inputs in the system prompt. Example : You will take as input the following arguments: {{Email_To}}, {{Customer_Email}}, {{Reply_Email_ID}}
- 10. 11 Context Grounding – index – Giving the Agent Memory Using indexes in agents Indexes give agents access to permissioned knowledge bases. This helps agents reason using business- specific data. You can upload: • Excel files • PDFs • Json • Docx • Txt Example: Shipping Cost Excel → agent uses them to validate inputs.
- 11. 12 Context Grounding – Semantic Or Structured ? Semantic What does it do? When you ask the agent a question, it finds and uses the most relevant text chunks based on meaning.Use this if your data source contains text content (e.g., PDFs, Word documents, emails, descriptions). Extra Settings: Relevance Score Threshold A value between 0 and 1. The higher the value, the more relevant and accurate the results. (But if it’s too high, it may return nothing.) Max Results Generated Sets how many text pieces will be used in the response. More results = higher token cost for the LLM. Structured Select this option for structured queries if you are using tabular data sources. Supports CSV data format.
- 12. 13 •Inputs and Outputs (Arguments) Define Arguments Supported Argument Types : String, Number, Integer, Boolean, Object, Array (For Table) • Inputs: what info the agent needs. • Outputs: what it returns (result, reason, decision). • Example: • Input: Invoice ID, Total Cost • Output: Status, Explanation
- 13. 14 Add Escalation – When Agent Gets Stuck Content: Define when and how the agent escalates to a human (Action Center, email). Useful for: Missing data Unexpected error Approval required Url for download app template : https://marketplace.uipath.com/listings/uipath-apps- templates
- 14. 15 Evaluations When you're building an agent, the goal is to make it reliable—something you can trust to give the right output consistently. Evaluations help you figure out if your agent is doing a good job or if it needs improvement. When to create evaluations Create evaluations once the arguments are stable or complete. That also means your use case is established, and the prompt, tools, and contexts are finalized. If you modify the arguments, you need to adjust your evaluations accordingly. To minimize additional work, it's best to start with stable agents that have well-defined use cases. You can export and import evaluation sets between agents within the same organization or across different organizations. As long as your agent design is complete, you can move evaluations around as needed without having to recreate them from scratch.
- 15. 16 Evaluators There are 3 evaulators ; a.LLM-as-a judge: Semantic Similarity – Creates your own LLM-based evaluator. (Default) b.Exact match – Checks if the agent output matches the expected output. c.JSON similarity – Checks if two JSON structures or values are similar. Target Output Fields ; •Root-level targeting (* All): Evaluates the entire output. •Field-specific targeting: Assesses specific first-level fields. Use the dropdown menu to select a field. The listed output fields are inherited from the output arguments you defined for the system prompt.
- 16. 17 What can Agents do in automation? Semantic data operation (data interpretation) Managing and optimizing data semantically to support decision-making and operational efficiency Organize data (documents, email etc..) Into predefined categories Pertain to pulling specific, relevant information from larger datasets or unstructured data (documents, email etc..) Detect and correct (or remove) corrupt or inaccurate records from a dataset Determine multiple data are matched semantically Explore and examine data within a dataset or collection to locate specific information or data elements Ensure that data or input meets specific criteria or standards Documents Invoice Po Receipt Agent Agent Agent Classification Extraction Formatting Matching Searching Validation Du Cm Agent Subject: address change policy no. 1863325 hi aaron, could you please update the address on the above policy to 20 W 34th st., New york, NY 10001, USA. Policy no: 1863325 Address line: 20 W 34th st., New york, NY Address line: 10001 Email 1600 pennsylvania ave northwest, washingtom, D.C 20500 1600 pennsylvania ave NW, washington, DC 20500 Subject: address change policy no. 1863325 Hi aaron, could you please update the address on the above policy to 20 W 34th st., New york, NY 10001, USA. Policy no: 1863325 Address line: 20 W 34th st., New york, NY Address line: 10001 Agent 1600 PENNSYLVANIA AVE NORTHWEST, WASHINGTON, DC 20500 1600 PENNSYLVANIA AVENUE NW, WASHINGTON D.C. 20500 MATCH MISMATCH Agent All third-party software must encrypt all personal data Pass Fail Contract Policy Signing a new contract with a software vendor Policy document archive Context • Data protection & privacy • Information security • Compliance & regulatory • Vendor management • Intellectual property • Business continuity & disaster recovery • Ethics and conduct Identified policy document list Du Cm Supported by communications mining Cm Supported by document understanding Du
- 17. 18 What can Agents do in automation? Natural language processing Understand and interact with human language and output meaningful sentence Llm 1600 pennsylvania ave northwest, washington, DC 20500 1600 pennsylvania avenue NW, washington D.C. 20500 Match Mismatch Interacting with humans in natural language, including responding to queries, providing information Identifies the sentiment expressed in a piece of text, helping to gauge public opinion, customer sentiments Communication Sentiment detection Condensing text or information into more concise formats while retaining the essential context or insights Converting text from one language to another while retaining the original meaning Summarization Translation Agent Agent Agent Agent Loan type: mortgage Loan amount: $ 950,000 Interest rate: 3.41% We are pleased to acknowledge your application for a loan amount of $950,000. Based on your profile and history with us, you may qualify for rates as low as 3.41%. Web page Email Uipath announced new ai-powered automation features, including generative AI and specialized AI, to enhance business automation. Web page UiPath announced new ai-powered automation features, including generative AI and specialized AI, to enhance business automation. UiPathは、generative aiや specialized AIなど、AIを活用した新 たな自動化機能を発表し、ビジネス の自動化を強化した。 English Japanese I received my order and found the product to be defective (it won't power on). I am very disappointed. Email Cm Supported by communications mining Cm
- 18. 19 What can Agents do in automation? Reasoning Drawing conclusions or making predictions based on existing knowledge, simulating human logical thinking and problem-solving capabilities Providing new insights based on provided information and its underlying principles Analysis Providing next actions strategies based on provided context, insights, trends, or patterns Suggestion Agent Crm Data base Financial statement Several business opportunities are identified based on the provided data #opportunity 1 ……………………… #opportunity 2 ……………………… Agent Crm Data base Financial statement Here is the generated account strategy based on the provided data #account strategy ………………………
- 19. 20 Rule Based - DEMO Mail gelir Completed Regex ile String kütüphanesi ile rule based olarak çıkartılır. Vendor Id, Departure date, Loading and Unloading port değerlerine göre filtrelenir. Maildeki Sea Freight, Local Cost ve Local Fixed değerleri hesapalnır, hesap tutarsa işlem biter. Excel aktiviteleri ile Excel okunur ve Datatable nesnesine çevirilir
- 20. 21 Agentic Process - DEMO Mail comes (Event trigger – Email Received) Completed Automation Calculation Agent extracts all values which are needed from mail body Try to find proper vendor id, Sheet name, loading port and unlıading port to get Sea Freight , local fixed, local charges Agent Look up For price List Agent Analyze & Extract Agentic Orchestration – Process Agent Human Calculateand compare Price List prices with e mail prices.
- 21. 22 AGENTIC ORCHESTRATION - MAESTRO
- 22. 23 Continuously improve processes and agents MAESTRO Orchestrate work across multiple systems Execute and manage processes at scale 3 Key Benefits For Maestro
- 23. 24 MAESTRO Example View of Maestro
- 24. 25 DEMO TIME 60 dk • Agent Structure with all components • Context Grounding • Escalation with UiPath Apps Structure • Agent Evalualtion Sets • Publish Agents To Orchestrator • Run Agent From UiPath Studio • Call Agents and Rpa’s from Maestro
- 25. 26 Dinlediğiniz için teşekkür ederim. 10 dakika Soru & Cevap oturumu