Paul Iusztin

🤖 Senior Machine Learning Engineer • MLOps • Contractor ~ Helping startups engineer production-ready ML/AI systems.

I'm the author of the bestseller LLM Engineer's Handbook, lead instructor of the Agentic AI Engineering course, founding AI Engineer of a San Francisco start-up, and obsessed with making knowledge accessible through AI.

With over 10 years of experience in AI and software and 20 apps shipped, I teach AI Engineering as I wanted to at the beginning of my career. End-to-end. From idea to production. From data collection to deploying, monitoring and evaluation. With a focus on AI principles, software patterns and infrastructure systems that will thrive in a future dominated by AI coding tools.

My ultimate goal is to help other engineers escape PoC purgatory and 10x their AI Engineering skills.

.

If you want to become an AI Engineer pro, join Decoding AI for exclusive content on designing, building, and shipping AI software that works: 

🗞️ https://www.decodingai.com

See more:

🌐 Homepage: https://www.pauliusztin.ai
📚 LLM Engineer's Handbook: https://www.amazon.com/LLM-Engineers-Handbook-engineering-production/dp/1836200072/
🤖 Agentic AI Engineering Course: https://academy.towardsai.net/courses/agent-engineering?ref=b3ab31

.

✉️ For collabs, contact me at: p.b.iusztin@gmail.com

Senior Machine Learning Engineer

🟡 Scalable ML batch infrastructure for continuous training & inference

The client, a generative AI leader, struggled with manual training and calling their ML models. As a tedious and unscalable process, they needed to automate their ML processes completely and integrate them with their current web platform.

➞ 𝐃𝐞𝐬𝐢𝐠𝐧𝐞𝐝 and 𝐢𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭𝐞𝐝 a modular 𝐏𝐲𝐭𝐡𝐨𝐧 𝐩𝐚𝐜𝐤𝐚𝐠𝐞 that composes the ML model required steps into 3 main batch pipelines that can easily be automated: feature, training, and inference, 𝐞𝐥𝐢𝐦𝐢𝐧𝐚𝐭𝐢𝐧𝐠 a 𝐭𝐞𝐝𝐢𝐨𝐮𝐬 𝟐-𝐡𝐨𝐮𝐫 𝐦𝐚𝐧𝐮𝐚𝐥 𝐩𝐫𝐨𝐜𝐞𝐬𝐬/𝐯𝐢𝐝𝐞𝐨.
➞ 𝐃𝐞𝐩𝐥𝐨𝐲𝐞𝐝 a scalable & cost-effective 𝐛𝐚𝐭𝐜𝐡 𝐚𝐫𝐜𝐡𝐢𝐭𝐞𝐜𝐭𝐮𝐫𝐞 and 𝐂𝐈/𝐂𝐃 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 on top of AWS ECS & SQS that scales in and out based on the number of jobs, 𝐬𝐚𝐯𝐢𝐧𝐠 𝐭𝐡𝐨𝐮𝐬𝐚𝐧𝐝𝐬 𝐨𝐟 𝐝𝐨𝐥𝐥𝐚𝐫𝐬/𝐦𝐨𝐧𝐭𝐡.
➞ 𝐒𝐚𝐯𝐞𝐝 𝐦𝐨𝐧𝐭𝐡𝐬 𝐨𝐟 𝐟𝐮𝐭𝐮𝐫𝐞 𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭 by using good software practices to make the Python module & batch cloud infrastructure flexible and composable.
➞ 𝐁𝐮𝐢𝐥𝐭 a 𝐯𝐨𝐢𝐜𝐞 𝐬𝐲𝐧𝐭𝐡𝐞𝐬𝐢𝐬 𝐏𝐨𝐂 in ~1 week required to 𝐜𝐥𝐨𝐬𝐞 the 𝐝𝐞𝐚𝐥 with one of the 𝐜𝐮𝐫𝐫𝐞𝐧𝐭 𝐜𝐥𝐢𝐞𝐧𝐭𝐬.

⚫️ Big data ingestion & archiving pipeline

The client wanted to expand its platform features by securely and safely ingesting, preprocessing, and storing TB of data into their platform.

➞ 𝐂𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐞𝐝 to the 𝐝𝐞𝐬𝐢𝐠𝐧 and 𝐢𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧 of a 𝐝𝐚𝐭𝐚 𝐢𝐧𝐠𝐞𝐬𝐭𝐢𝐨𝐧 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 that supports 𝐢𝐧𝐠𝐞𝐬𝐭𝐢𝐧𝐠 >𝟓 𝐓𝐁 of data in 𝐨𝐧𝐞 𝐠𝐨 in a reliable, fast, and secure way.
➞ 𝐈𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭𝐞𝐝 a scalable 𝐝𝐚𝐭𝐚 𝐩𝐫𝐞𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠 𝐩𝐢𝐩𝐞𝐥𝐢𝐧𝐞 that supports multiple media types (images, videos, audio, 3D objects) and creates small preview files that are cheap to transfer, store and visualize. The 𝐨𝐫𝐢𝐠𝐢𝐧𝐚𝐥 𝐟𝐢𝐥𝐞𝐬 are 𝐚𝐫𝐜𝐡𝐢𝐯𝐞𝐝 on an affordable storage solution 𝐫𝐞𝐝𝐮𝐜𝐢𝐧𝐠 𝐜𝐨𝐬𝐭𝐬 𝐛𝐲 𝐮𝐩 𝐭𝐨 𝟖𝟏%.

I built a system that turns a single idea into a publish-ready technical article.

Here's a mini masterclass on how (so you can copy):

The system consists of 2 parts:
 
• Nova (research agent): exploratory, steerable
• Brown (writing workflow): deterministic, auditable

𝟭) 𝗡𝗼𝘃𝗮: 𝗦𝘁𝗲𝗲𝗿𝗮𝗯𝗹𝗲 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗔𝗴𝗲𝗻𝘁 (𝗲𝘅𝗽𝗹𝗼𝗿𝗮𝘁𝗶𝗼𝗻)

Goal: ingest context from messy sources, fill gaps, and produce a clean research artifact.

Inputs:
 
• Structured article brief (topic, audience, structure, constraints, candidate sources)

How it runs:
 
• MCP-first design so tools are portable across harness clients
• A prompt exposed as procedural memory to plan and glue all stages
• Tooling focused on ingesting context, not “thinking in tokens”

Core stages:
 
• Pull context from URLs, local files, GitHub repos, YouTube transcripts
• Run multi-round research loops (Perplexity for cited web search)
• Filter for trust and relevance
• Deep scrape the top sources
• Consolidate into a single handoff artifact: research.md

Output:
 
• A curated research document with organized notes and citations, ready for drafting

𝟮) 𝗕𝗿𝗼𝘄𝗻: 𝗥𝗲𝗹𝗶𝗮𝗯𝗹𝗲 𝗪𝗿𝗶𝘁𝗶𝗻𝗴 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄 (𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻)

Goal: turn research into a consistent article, with quality gates and checkpoints.

Why a workflow, not an agent?

Because writing quality demands:
 
• Repeatability
• Checkpoints
• Clear stage boundaries
• Traceability when something is off

How it runs:
 
• LangGraph for stateful orchestration, resumability, and auditability
• Leveraging tools as deterministic workers instead of exposing them to agents

Core stages:
 
• Load context: guideline + research.md + writing profiles
• Generate required assets in parallel (diagrams, images, tables when needed)
• Draft the article
• Run review and rewrite loops (reflection scores, style judge, revision triggers)
• Support human edits without breaking the system’s rules

Outputs:
 
• Final article
• Generated assets + metadata
• Reflection scores and traces (so you debug quality instead of guessing)

Here’s the gist…

If you promise “production-grade writing”, you need exploration for research and durability for execution.

This design choice is the difference between a cool demo and a system you can ship.

Want to learn how to build it out yourself?

We break this down step-by-step in the Agentic AI Engineering course (built with Towards AI, Inc.):

Enroll here: https://lnkd.in/dtxZuZpb

P.S. We’ve opened up another 100 seats for $499. The last batch sold out in under 14 days → get in early to lock in our best offer.

Senior AI Engineer • Founder @ Decoding AI • Author @ LLM Engineer’s Handbook ~ I ship AI products and teach you about the process.

Established & led enterprise-wide Data Analytics & BI CoE:
- Built on Azure cloud native DWH across domains.( Azure Synapse as DWH, Spark for API integration and ADLS Gen 2 for data storage and Power BI as data visualisation and front end transformation and Power Platform for automation and self service enablement)
- Enabled touch-less month end financial run for P&L generation at enterprise level with an efficiency gain of 50%
- Data & BI Hub for the entire organization consisting of more than 3 TB of data and growing constantly

Pioneered Generative AI for democratized insights:
- Spearheaded "CFO Copilot" initiative, for personalized insights from diverse datasets using RAG , prompt engineering and database agents
- leveraged Azure Open AI services , Langchain orchestration, Function and Tool calling for more deterministic response.
- Internal dataset derived from DWH on Azure Synapse 
- External dataset ingested using Yahoo Finance API and earnings call artefacts and 10k/10Q artefacts ingested in Azure Data Lake Gen2 ( using hierarchical namespace for better access control)
- Azure AI search as vector DB and RAG application (hybrid search ) to read densely populated excel files or tables based pdf documents using markdown options.

- Set up Generative AI R&D, enabling faster solution & IP development.
Championed innovation & efficiency:
- Led enterprise DevOps practice & implemented GitHub Enterprise on Cloud.
- Hosted knowledge sessions for peer-to-peer learning & engagement.
Strategic Business Transformation:
- Managed Business Architecture analysis for customer-centric digital transformation.
- Designed solutions based on process complexity, digital maturity, & market demand.
Coaching & Mentorship:
- Agile principles & SAFe framework coach for in-house teams & external audiences.
- Facilitated client workshops for customer journey maps & digital roadmaps.
Industry Recognition:
Speaker at Data and BI Forums, Business Analyst Conventions & Agile Forums.

Senior AVP - Transformational BI & Generative AI Leader | Democratizing Insights, Fueling Innovation

Espire Infolabs

•	Developed Customer Journey Framework as selling pitch ,establishing organisation credibility to facilitate Customer Journey Maps workshops with clients to design transformation roadmap. 
•	Awarded the Best Insurance Account Team by contributing in account mining of an insurance client by providing analytical insights in their current process as Insurance CoE Head
•	Developed PoC for Data driven KPIs like churn rate , click through , return customers , customer sentiment analysis etc on customer experience driven solution , used for client pitching. 
•	Pitching along with the sales team , technical architects and  offerings to clients for winning deals by articulating outcome oriented strategy while effectively managing the risks and solution / commercial feasibility of the proposal
•	Responsible for developing GTM Strategy as Presales Head enabling winning strategy by combining CJM framework with organisational capabilities and offerings.

Insurance CoE & Presales Head

Aviva Singapore(SQS Group)

•	Facilitating workshops with business and developers for requirement elicitation and story boarding
•	Product Backlog grooming leveraging use of epics, user stories and user story mapping for delivery of functional value to client leading to functional useful sprint releases.
•	Assessing and Assigning Story Points to each story and acting as act as the "front office" product owner to internal teams for requirement clarification  and functionality sign-off 
•	Planning and prioritizing for the delivery of useful and usable digital solutions releases in Ipads, meeting the product launch dates effectively.
•	Leading and managing UAT in an onshore-offshore delivery model 
•	Negotiating and managing stakeholder expectations

Digital Business Analyst cum Scrum Master

SAP ( Serole Technologies)

•	Responsible for managing requirements and solution design to replace the legacy system (AS400) with SAP Core Insurance Platform, for a leading General insurer in Singapore, MSIG, by redesigning the business processes and leveraging offshore development and onshore engagement, achieving improved operational efficiency and service quality by more than 60% 
•	Through efficient Product Backlog grooming and user story point estimation, was able to deliver sprints deliverables on time, resulting in improved efficiency by around 50%
•	Have successfully led the migration of over 150 products configuration in the new solution from the legacy system (AS400) 
•	Facilitated workshops with both business and technology team in identifying, prioritizing and implementing requirements and solution assessment, in line with client's business model, strengths and mitigating risks that could potentially impact operating performance, thereby achieving win-win results
•	Successfully managed the project timelines and resource, working in a multi - location based team, making a significant contribution to the 3rd quarter best project team award winning from SAP.

Lead Business Analyst / Product Owner

•	Partner with senior management and stakeholders on leading enterprise change programs and organizational change initiatives to drive innovation and continuous improvements within the company , thereby being awarded for evangelizing Agile best practices.
•	Process Discovery, Modeling and Analysis for gaps between as-is and to-be, waste points and improvement opportunities in business processes for both cross-stream and single stream, enabling innovation and transformation (including digital transformation)
. Was the key member in the product innovation team , of which one the innovation, Clinical Decision Support System, designed based on Design Thinking , was acknowledged as "A great Idea" by CEO of EXL.
•	Lead sessions with both business and technology stakeholders, senior and relevant, for change requirement elicitation, change management strategy and readiness assessment.
•	Change Scope definition by developing business process models that can be simulated, analyzed and executed by business.
•	Identify opportunities for automation for increasing process efficiency and effectiveness , resulting in 30 - 40% of FTE saving and 40-50% improvement in turn-around time.
•	Conduct Market Research to identify and innovate best fitted solution. 
•	Design solutions, both IT and non-IT to meet the business needs, in alignment to Target Operating Model.
•	SWOT and Cost Benefit Analysis of the identified solution.
•	Developing Change Management Strategy and Plan.
•	Streamlining of processes with the adoption of techniques such as Agile Principles, Six Sigma and Design Thinking
•	 Participated both in Scrum based  and traditional project management
•	Assist with the creation of testing strategy, scenarios creation, testing, and training for new / enhanced systems. 
•	Project Scoping and Estimation

Manager - Business Process Management

Max Bupa Health Insurance Company Ltd.

Conducting workshops and brainstroming sessions to elicit, analyse and identiy  the actual  business user requirement and business process mapping
Perform Gap Analysis , Data Assessment (around KPI/metrices) and Issue Tree Development
Development and review  of software specification document s such as Functional Specification Document
To train business users on the software product
To be an integral part in creating test plan and test cases
To manage the User Acceptance Testing  and Change Management phase

Business Analyst-IT

Steria India Ltd

Business Process Modelling,requirement analysis workshops and quality analysis  for Process Optimisation initiatives 
Conducted risk analysis, undertook projects on insurance software products advancements and underwriting developments   
Assessment of the insurance proposals related to Life and mortgage policies and taking appropriate underwriting decisions      
Acted as a facilitator of various meetings with clients and cross team members for process improvement initiatives and prepared high level requirement documents
Have been involved in mentoring and team management activities
Worked closely with QA team in identifying and fixing the bugs and resolving any issues

Assistant manager Specialist

Genins India Ltd,

Managing Health Claims team of 10 people
MIS Reports and Dashboards for status updates
Process Improvement Initiatives 
Claims Grievance Handling and Pre-Authorization process management

Claims Supervisor

TTK HealthCare Services Ltd.

Processing of Claims and Pre-authorization
Network Management with Providers 
Customer Grievance Management

Claims Executive

At EXL, my role as Senior AVP encompasses not only the leadership of our Enterprise Data Analytics and BI Center of Excellence but also pioneering generative AI to revolutionize insights across the enterprise. With a firm grasp on Microsoft Power Platform , Power BI, Azure Synapse, Data Engineering,Azure Cognitive/ AI Services and a track record in business process improvement(Six Sigma Green Belt certified) , my expertise lies in harnessing data to fuel innovation and enhance operational efficiency. 

The conversational BI initiative powered by Generative AI ,stands as a testament to our team's success in personalizing data insights, achieved through diligent application of data engineering and AI best practices. In collaboration with peers, we've elevated the enterprise's DevOps practices, fostering an agile environment conducive to swift and impactful digital solution delivery. My commitment to knowledge sharing and empowerment is reflected in the numerous sessions I've hosted, aimed at peer learning and engagement.

Koyelia Ghosh Roy

Most #AgenticAI Delivery teams I meet can build a killer demo in a week… and then get stuck for months trying to ship it safely.
The two blockers are almost always: Security + Evaluation.
Specifically:
**How do we ensure an agent isn’t leaking PII (credit cards / SSNs) or getting manipulated by prompt injection payloads embedded inside documents? **
Prompt injection is a real class of attacks where adversarial instructions are disguised as normal input, with the goal of overriding system intent. 
So I built a **Red Team evaluator workflow** for an #AIInvoiceAgent using custom MLflow metrics—designed to run inline (not weeks later in a spreadsheet). 
#MLflow 3.9.0 is putting serious weight behind observability + evaluation, which makes this pattern feel like “table stakes” for production Agentic AI solution.
Key takeaways (what I’d recommend every team adopt):
-Use deterministic evaluators wherever possible
-Regex / rules-based PII checks are fast, cheap, and explainable—no LLM needed for obvious patterns. 
-Use an LLM judge for the complex stuff
-Prompt injections are often obfuscated and contextual—this is where an LLM-as-a-judge adds real value. 
-Make “policy gates” non-negotiable
-If any security score trips → block or redact automatically. Don’t rely on “we’ll review logs later.”
-Log an audit trail like it’s a security system (because it is)
-Tracking URI to SQLite is a simple way to get structured storage for runs/metrics locally (and it scales to a tracking server later).

If you’re deploying agents that touch invoices / HR docs / medical forms:
Do you run security evals inline today—or only after incidents happen?

#MLflow #GenAI #LLMOps #MLOps #AIsecurity #PromptInjection #Evaluation #Observability #3AI #WIC 

Jules Damji Carly Akerly

Real-time Red Team Evaluation for an AI Invoice Agent: Deterministic PII detection + LLM-based prompt-injection judging, enforced by a policy gate and

Open article: From Agentic AI Demos to Production: Real‑Time Red Team Evaluation with MLflow 3.9.0 by Koyelia Ghosh Roy • 3 min read

From Agentic AI Demos to Production: Real‑Time Red Team Evaluation with MLflow 3.9.0

I’m attending the 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐂𝐨𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞 by Data Science Dojo this 𝐀𝐩𝐫𝐢𝐥 𝟔–𝟏𝟎, 𝟐𝟎𝟐𝟔.
 
This 𝐅𝐑𝐄𝐄 virtual conference brings together AI professionals and innovators to explore the rapidly evolving world of Agentic AI.
 
👉Join thousands from around the globe: https://hubs.la/Q03FZGwN0
 
#agenticaiconference #datasciencedojo

**90% of Agentic AI Projects Fail Here (And It's Not the Model)**
In enterprises, the hardest part of Agentic AI is not the model.
It's encoding how decisions are actually made — under policy, pressure, and scrutiny.
Most critical processes are:
-Poorly documented
-Exception-heavy
-Judgment-driven
-Highly auditable and regulated

In reality, the decision logic lives in experienced operators and fragmented SOPs.
If you don't extract that logic, your agents won't scale beyond demos.
Across multiple deployments in financial services and operations, this disciplined approach can reduce agent escalations by 40% and cut audit prep time in half.
Here's the framework I use in production:

1️⃣ Mine Decisions, Not Just Workflows
-Don't start with swimlanes. Start with decisions.
-Replay historical cases (wins, losses, escalations, audit hits)
-Capture why a decision was made, not just what step happened next
-Annotate edge cases that forced humans to override the process
-You're building a decision graph, not a process diagram.
2️⃣ Make Judgment Explicit
-Turn tacit judgment into structured logic.
-For each key decision, extract:
-Risk triggers
-Materiality thresholds
-Escalation conditions
-Evidence requirements
This becomes the playbook your agents operate within — not a vague prompt.
3️⃣ Design Confidence Bands
-Agents should not be "all or nothing."
-Define clear bands:
-High confidence → Auto-execute within limits
-Medium confidence → Recommend with rationale
-Low confidence → Escalate with context and options
This gives you controlled autonomy instead of uncontrolled automation.
4️⃣ Engineer Auditability From Day 1
-If it can't explain itself, it can't operate at scale in a regulated enterprise.
-Every agent action should log:
-Inputs and data sources used
-Rules / heuristics applied
-Confidence score and version of the playbook
-Override history and human approver
This is how you satisfy risk, compliance, and regulators and keep trust high.
💡 The Bottom Line:
Smart agents are not prompt-driven.
They are **playbook-driven.**
This discipline of structured playbook engineering is foundational to how I approach Agentic AI in AION™ — where agents operate inside a defined decision and governance network, not just ad-hoc prompts.
🎯 What's Your Biggest Blocker?
If you're leading AI transformation, what's stopping your agents from moving beyond pilots?
Comment below:
📊 Extracting decision logic
⚖️ Handling edge cases & exceptions
🔒 Audit/compliance requirements
💼 Executive buy-in & trust

#AgenticAI #EnterpriseAI #AIStrategy #AIGovernance #DigitalTransformation #AITransformation #3AI #WIC #WIT

taplio

The Battle

Paul Iusztin

Koyelia Ghosh Roy

Followers

Posts

Engagements

Likes

Performance ScorePaul Iusztin's LinkedIn Performance

Performance ScoreKoyelia Ghosh Roy's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Paul Iusztin

Koyelia Ghosh Roy

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Paul Iusztin's LinkedIn Performance

Performance ScoreKoyelia Ghosh Roy's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScorePaul Iusztin's LinkedIn Performance