AI that ships.
Isuru (IG) Chathuranga
AI/ML Engineer
I Ship AI Systems That Actually Work in Production. Building production-first AI systems, agentic workflows, and automation pipelines. Not just models, but deployed systems that scale and self-heal.
Core Tech Stack
About Me
Shipping AI That Works in Production
Most AI engineers can train models. I deploy them at scale, monitor performance, and fix them when they break. That's the difference.
Final-year at University of Colombo (IT & Management Hons | Applied Statistics & Data Science), building production AI systems where statistical rigor meets MLOps and AIOps.
Currently Technical Assistant on the BuildrLabs.ai Agentic AI Bootcamp, supporting engineers through labs covering LLMs, RAG, LangGraph/CrewAI agents, custom MCP servers, Kubernetes, and full production observability with Prometheus, Grafana, and OpenTelemetry.
I focus on AI systems that actually work in production, not just notebook demos. From multi-agent architectures and n8n automation pipelines to MLOps with drift detection, I build systems that handle real traffic and self-heal when things break.
Open source contributor to LangChain, PyTorch, Hugging Face Transformers, and MLflow, helping thousands of ML engineers deploy reliable systems at scale.
Education
BSc IT & Management (Hons)
University of Colombo
Final Year • 2023-2027BIT, UCSC (2023-2026)
Location
Colombo, Sri Lanka
Remote Ready
Availability
Open to Remote Opportunities
Full-time • Freelance • Collaborations
Focus Areas
LLM Engineering • Multi-Agent Systems
MLOps/AIOps • Production ML
Services
What I Build for Clients
Agentic AI Assistants
MULTI-CHANNEL CHATBOTS
AI assistants that handle WhatsApp, web chat, and email 24/7. Custom RAG over your private docs, human-in-the-loop checkpoints, and analytics built in.
Voice AI Agents
24/7 CALL HANDLING
Full-duplex voice agents on Twilio + Deepgram + OpenAI with sub-second latency. Handle bookings, FAQs, and routing in English, Sinhala, or Tamil.
Workflow Automation
n8n • MAKE • ZAPIER
Production automation pipelines that connect LLMs to real-world tools. Payment verification with bank reconciliation, Meta Ads decisioning agents, and multi-step approvals with human checkpoints.
Custom ML & Knowledge Systems
RAG • GRAPHRAG • DEEP LEARNING
Custom models trained on your data: churn prediction, fraud detection, medical imaging. Plus RAG/GraphRAG knowledge engines that let your team query private data in plain language.
Tech Stack
AI/ML Technology Arsenal
Production ML tools and frameworks I use to build intelligent systems
44+
Technical Skills
16
Production Ready
5
Soft Skills
★★★★ Proficient - Production experience | ★★★ Intermediate - Active learning | ★★ Familiar - Working knowledge
Showing 46 of 46 technologies
PyTorch
Proficient
Primary DL Framework
PEFT (LoRA)
Intermediate
Model Fine-Tuning
LangChain
Proficient
Multi-Agent Orchestration
LangGraph
Proficient
Stateful Agent Workflows
CrewAI
Intermediate
Multi-Agent Framework
MCP
Intermediate
Model Context Protocol
Transformers
Intermediate
Pre-trained Models
OpenAI API
Intermediate
GPT Integration
Anthropic Claude
Intermediate
Claude API
Groq
Proficient
Fast LLM Inference
Qdrant
Proficient
Vector Database
Twilio
Intermediate
Voice AI Telephony
Deepgram
Intermediate
Speech-to-Text
FastAPI
Proficient
ML API Development
Docker
Proficient
ML Containerization
Kubernetes
Intermediate
Model Orchestration
MLflow
Intermediate
Experiment Tracking
LangSmith
Intermediate
LLM Observability
Prometheus
Intermediate
Metrics & Monitoring
Grafana
Intermediate
Observability Dashboards
Git
Proficient
Version Control
GitHub
Proficient
Code Collaboration
Redis
Familiar
Feature Store/Cache
GCP (Vertex AI)
Intermediate
ML Deployment
AWS (Bedrock)
Intermediate
ML Services
Azure OpenAI
Familiar
ML Platform
Python
Proficient
Primary Language
scikit-learn
Proficient
Classical ML
Pandas
Proficient
Data Manipulation
NumPy
Proficient
Numerical Computing
Jupyter
Proficient
Interactive Notebooks
XGBoost
Intermediate
Gradient Boosting
R
Intermediate
Statistical Analysis
PostgreSQL
Intermediate
Data Storage
Neo4j
Intermediate
Graph Database
Apache Kafka
Intermediate
Event Streaming
Apache Airflow
Intermediate
Workflow Orchestration
VS Code
Proficient
Development
n8n
Proficient
Workflow Automation
Make
Familiar
Visual Automation
Zapier
Familiar
App Integrations
Problem Solving
Proficient
Analyzing complex problems
Technical Communication
Proficient
Conveying complex concepts
Collaboration
Proficient
Cross-functional teams
Adaptability
Proficient
New tech & environments
Continuous Learning
Proficient
Industry trends
Continuously expanding expertise through production projects and open source contributions
Portfolio
Production ML Systems
AI systems deployed at scale with real business impact
PersonaPlex: Full-Duplex Voice AI Agent
Production-ready voice AI agent that answers real phone calls with sub-second latency.
Challenge
Building a voice AI agent capable of handling real-world phone calls with natural interruptions and ultra-low latency.
Solution
Integrated Twilio, Deepgram for fast STT with end-of-speech detection, OpenAI for logic, and Edge TTS to create a highly responsive, customizable voice agent.
Key Results
- Sub-second latency for voice responses
- Robust speech-to-text with end-of-speech detection
- 100% customizable personas
- Full-duplex conversation handling
Real-Time AI Research Agent
Agentic system with autonomous web search, custom RAG, and automatic source citations.
Challenge
Needed an autonomous research assistant that can effectively search the web, compile information reliably, and provide accurate citations.
Solution
Built a LangGraph and FastAPI-based agentic system with custom RAG, robust error handling, and lazy initialization for seamless deployment.
Key Results
- Autonomous web search capabilities
- Custom RAG with automated source citations
- Robust error handling and validation
- Hosted on Hugging Face Spaces
Cost-Aware Fraud Detection System
Production ML system optimized for business metrics (cost of false positives vs. false negatives)
Challenge
Built complete ML pipeline from feature engineering to production deployment that optimizes true business costs.
Solution
Developed end-to-end system with feature engineering, model training, evaluation tracking, deployment, and continuous monitoring.
Key Results
- Optimization for business metrics and costs
- End-to-end ML pipeline from features to production
- 92% test coverage and MLflow tracking
- Production deployment with monitoring
Sri Lankan Market Intelligence Agent
AI-powered market intelligence system analyzing Sri Lankan economic indicators
Challenge
Needed real-time market intelligence system to analyze complex economic data from multiple sources with automated insights.
Solution
Built multi-agent LangChain system with custom RAG architecture, safe calculator tools, and intelligent market analysis capabilities.
Key Results
- Production-ready LangChain agent implementation on GCP
- Real-time processing of economic indicators
- Monitoring and automated failover
- Intelligent query handling with custom RAG
ML Monitoring & Drift Detection System
Production-ready ML monitoring with automated drift detection and real-time dashboard
Challenge
Production ML systems need automated drift detection, intelligent alerting, and self-healing capabilities.
Solution
Built comprehensive MLOps pipeline with model drift detection, automated retraining triggers, real-time dashboard, and performance monitoring.
Key Results
- Automated model drift detection and alerts
- Real-time monitoring dashboard
- Self-healing pipelines with alerts
- Docker-ready with CI/CD integration
Telco Churn Prediction Pipeline
End-to-end ML system from feature engineering to production deployment
Challenge
Needed a validated ML business solution capable of predicting customer churn securely and proactively.
Solution
Developed a complete PyTorch-based ML pipeline covering feature engineering, model training, and production deployment.
Key Results
- End-to-end ML pipeline creation
- Business-validated predictions driving decisions
- Production deployment with automated CI/CD
- Comprehensive MLOps integration
Multi-Modal Crypto Intelligence System
AI-powered trading signal generator with real-time sentiment orchestration
Challenge
Crypto markets require real-time sentiment analysis from multiple sources with sub-second latency for actionable trading signals.
Solution
Developed multi-agent system with Whale Tracker, Orderbook Analyzer, Sentiment Analyzer, and LLM Coordinator for comprehensive analysis.
Key Results
- Real-time trading signal generation with sub-2s latency
- Multi-agent architecture with specialized agents
- AIOps integration for anomaly detection
- Docker-ready deployment with WebSocket support
Open Source Contributions
Contributing to leading AI/ML frameworks used by thousands worldwide
Challenge
Improving production ML tools and frameworks for the broader ML community.
Solution
Active contributions to LangChain, PyTorch, Hugging Face Transformers, and MLflow. Focus on production-ready implementations and documentation.
Key Results
- Contributions to LangChain, PyTorch, Hugging Face, MLflow
- Production-ready examples and tutorials
- Code quality and documentation improvements
- Supporting thousands of ML engineers
Adaptive Multi-Adapter LLM Framework
Production-ready LoRA/QLoRA fine-tuning with intelligent routing and self-evaluation
Challenge
Need scalable framework for fine-tuning LLMs on domain-specific tasks with intelligent adapter selection and automated quality assessment.
Solution
Built multi-adapter system with BERT-based router network for dynamic adapter selection, multi-metric self-evaluation, and uncertainty quantification using MC Dropout.
Key Results
- Learned router network with Gumbel-Softmax selection
- 4 specialized adapters: Reasoning, Code, Creative, Analysis
- Multi-metric self-evaluation (relevance, coherence, factuality)
- Uncertainty quantification with MC Dropout
Preview Coming Soon
MediAgent SL
Multilingual agentic healthcare navigator for Sri Lanka
Challenge
Public healthcare system serves 22M+ people with no automated triage. OPD patients face 4–6hr waits without pre-screening in multiple languages.
Solution
A 7-agent hierarchical system on LangGraph 1.0 with MCP for FHIR R4 clinical data, offering trilingual voice-based medical intake.
Key Results
- Multi-Agent system with trilingual support (Sinhala, Tamil, English)
- Scalable triage engine handling dengue outbreak demand spikes
- Graceful fallback states and pessimistic routing for emergencies
- Dual output: patient explanation & clinical JSON (ICD-10/LOINC)
Preview Coming Soon
NEXUS: Agentic GraphRAG Engine
Multi-modal knowledge engine fusing dense vector retrieval with knowledge graph traversal for multi-hop reasoning
Challenge
Flat vector search can't handle multi-hop reasoning across interconnected entities. GraphRAG traversal is needed to follow relationships through a knowledge graph.
Solution
Building a hybrid retrieval engine combining Qdrant vector search with Neo4j graph traversal, orchestrated by LangGraph agents, with Redpanda streaming and Dragonfly cache for low-latency queries.
Key Results
- Hybrid dense + graph retrieval for multi-hop reasoning
- Qdrant + Neo4j fusion with LangGraph orchestration
- Redpanda streaming + Dragonfly cache for low latency
- Multi-modal input support (text, structured data)
ITM Exam Portal
Production exam management platform live and actively used by the faculty cohort at University of Colombo
Challenge
Faculty needed a reliable, self-managed exam portal for their cohort without dependence on university-wide systems.
Solution
Built a full-stack web application with exam scheduling, submission handling, and result management, deployed and live for real users.
Key Results
- Live in production, actively used by faculty cohort
- End-to-end exam lifecycle management
- Self-managed deployment with stable uptime
- Built and shipped solo under real deadline pressure
AI-Powered Automation Workflows (n8n)
Production automation pipelines connecting LLMs to real-world business tools with human-in-the-loop checkpoints
Challenge
Businesses need LLM-driven automation that integrates with existing tools (WhatsApp, Slack, banking APIs, Meta Ads), not standalone chatbots.
Solution
Built a suite of production n8n workflows: multi-channel payment verification with WhatsApp intake, document processing, and bank reconciliation; plus a Meta Ads AI agent for automated campaign decisioning.
Key Results
- Multi-channel payment verification: WhatsApp intake → document processing → bank reconciliation → Slack confirmation
- Meta Ads AI agent for automated campaign analysis and decisioning
- Human-in-the-loop checkpoints at critical approval stages
- LLM orchestration with real-world tool integrations via n8n
Experience
Production AI/ML Journey
Technical Assistant, Agentic AI Bootcamp
Buildr Labs • Colombo, Sri Lanka (Hatch Works)
Technical Assistant on BuildrLabs.ai's 16-week Agentic AI Bootcamp (May 2026 Cohort), training engineers to ship production-grade agentic AI systems at Hatch Works, Colombo.
Key Projects:
- •In-person lab support every Saturday, debugging across LLMs, RAG pipelines, agent workflows, and MCP servers
- •Async Discord support throughout the week, unblocking participants on technical issues outside class hours
- •Reviewing project submissions and commit history alongside the instructor team
- •Curriculum: LangGraph/CrewAI agents, 4-tier memory MCP servers, Kubernetes, multi-cloud deployment (AWS Bedrock, Vertex AI, Azure OpenAI), and cognitive observability with Prometheus, Grafana, and OTel
Open Source Contributor
Open Source Community • Remote
Contributing to leading AI/ML open-source projects to advance production machine learning tools and frameworks. Focus on improving code quality, documentation, and real-world implementations across the LLM and MLOps ecosystem.
Key Contributions:
- •LangChain - Enhancing multi-agent system implementations, improving RAG architecture examples, and fixing bugs in core chain orchestration modules
- •PyTorch - Contributing to deep learning tutorials, test coverage improvements, and documentation clarity for production deployment patterns
- •Hugging Face Transformers - Adding model integration examples, improving tokenizer documentation, and optimizing inference code paths
- •MLflow & MLOps Tools - Enhancing production ML tracking capabilities, adding deployment examples, and improving monitoring integrations
ML Systems Top 3 Winner
Zuu Crew AI • Remote
🏆 Top 3 Winner out of 290 participants in the intensive 13-week 'Build Production-Ready Machine Learning Systems' bootcamp. Recognized for exceptional performance in building end-to-end production ML systems.
Key Projects:
- •Built end-to-end ML pipelines with MLflow integration for experiment tracking and model versioning
- •Implemented scalable data processing workflows using Apache PySpark and Airflow orchestration
- •Developed real-time streaming prediction pipelines with Apache Kafka
- •Deployed production-grade ML systems with monitoring, CI/CD, and automated workflows
- •Capstone: End-to-end streaming-to-prediction pipeline with production deployment
Freelance AI Engineer
Self Employed • Remote
Building production-ready AI/ML systems for diverse applications including market intelligence, predictive analytics, and multi-agent architectures. Specializing in LangChain-based systems, PyTorch deep learning, and end-to-end MLOps pipelines deployed on cloud infrastructure.
Key Projects:
- •AI-Powered Sri Lankan Market Intelligence Agent - Multi-agent RAG system analyzing economic indicators with automated monitoring on GCP
- •Telco Churn Prediction Pipeline - End-to-end ML system from feature engineering to production deployment with drift detection
- •Multi-Agent Crypto Market Analyzer - Real-time sentiment orchestration with sub-2s latency and AIOps integration
- •Deployed production ML systems on AWS, GCP, and Azure with monitoring and self-healing capabilities
BSc IT & Management Student
University of Colombo • Colombo, Sri Lanka
Final-year student specializing in Applied Statistics & Data Science, with focus on production ML systems. Building AI systems where statistical rigor meets MLOps and AIOps. Also pursuing BIT at University of Colombo School of Computing (Oct 2023 - Oct 2026).
Academic Focus:
- •Specialization: Applied Statistics & Data Science
- •Focus: Machine Learning Engineering & Production ML Systems
- •Research: Statistical Programming & ML Theory
- •Additional: BIT at UCSC (2023-2026)
Blog & Articles
Latest Insights & Technical Writing
Certifications
Validated AI/ML Expertise
20+ industry-recognized certifications in AI, ML, and production systems
20+
Certifications
8
Providers
5
Categories
Recent
Nov-Dec 2025
Showing 28 of 28 certifications
Oracle Cloud Infrastructure 2025 Certified AI Foundations Associate
Oracle
AI Engineer for Data Scientists Associate
DataCamp
Python Data Associate
DataCamp
AI Agent Fundamentals with Azure AI Foundry
Microsoft
Databricks Accredited Generative AI Fundamentals
Databricks
Databricks Accredited AI Agent Fundamentals
Databricks
Qdrant Essentials
Qdrant
Deep Learning with PyTorch: Image Segmentation
Coursera
Sentiment Analysis with Deep Learning using BERT
Coursera
Introduction to Deep Learning with PyTorch
DataCamp
Model Context Protocol (MCP): Hands-On with Agentic AI
LinkedIn Learning
Introduction to Model Context Protocol
Anthropic (Coursera)
Master SQL for Data Analysis
Packt (Coursera)
Fine-tuning Language Models (The LLM Course)
Hugging Face
Fundamentals of LLMs (The LLM Course)
Hugging Face
Working with Hugging Face
DataCamp
Securing Generative AI
Pearson
MLOps Essentials: Model Development and Integration
LinkedIn Learning
Learning Docker
LinkedIn Learning
Astronomer Certification DAG Authoring for Apache Airflow 3
Astronomer
Astronomer Certification for Apache Airflow 3 Fundamentals
Astronomer
Machine Learning with Python
IBM
The Nuts and Bolts of Machine Learning
Supervised Learning with scikit-learn
DataCamp
Unsupervised Learning in Python
DataCamp
Regression Models
Johns Hopkins University
Understanding Bias in AI
LinkedIn Learning
Portfolio Optimization using Markowitz Model
Coursera
Get In Touch
Let's Build Production AI Together
Open to remote AI/ML roles, freelance projects, and collaborations on production systems
Currently seeking:
- •Remote AI/ML Engineering roles
- •LLM engineering & multi-agent systems
- •MLOps/AIOps infrastructure positions
- •Production ML pipeline development
- •Open source collaborations
isuruigic@gmail.com
Phone
+94 77 050 0424
Location
Colombo, Sri Lanka
Remote Ready • Available Globally
Response Time
Within 24 hours
Usually faster for interesting projects