| 🌐 Parameter | ⚡ Operational Specification |
|---|---|
| Full Name | Lasitha Dilshan Thilakarathna |
| Core Specialization | AI Engineer & Generative AI Architect |
| Experience Depth | 5+ Years (Full-Stack & Cloud ➔ Production AI Systems) |
| Primary Accolade | 🏆 1st Place Winner — Virtusa Agentic AI Hackathon |
| Architectural Focus | Autonomous Agents, Enterprise RAG, Vector Search, FastAPI Microservices |
| Location & Timezone | Bandaragama, Sri Lanka 🇱🇰 (UTC+05:30) |
| Collaboration Status | 🟢 Open to high-impact AI Engineering roles, consulting & research |
Spearheaded and architected the Agentic AI Travel Assistant, competing against enterprise engineering teams. Built autonomous multi-agent task delegation pipelines using Google ADK and OpenAI LLMs for dynamic itinerary planning and automated execution.
- 🎖️ Certified GenAI Assisted Engineer — Virtusa
- 🎖️ Career Essentials in Generative AI — Microsoft & LinkedIn
- 🎖️ Custom GenAI Enterprise Pathway — Virtusa
- 🎖️ GenAI Hackathon Certificate of Excellence — Virtusa
| Product / System | Description | Status | |
|---|---|---|---|
| 🏆 | Agentic AI Travel Assistant | Autonomous multi-agent travel planning system featuring dynamic goal decomposition, real-time itinerary orchestration, and automated execution workflows. | 1st Place Champion |
| 🔬 | Deep Research Agent | Autonomous web crawling, multi-source evidence extraction, factual cross-verification, and structured research report synthesis. | Open Source |
| 🤖 | Career-Ops | Local CLI agent that scans job portals, synthesizes candidate experience, scores listing fit (A-H), and tailors ATS-compliant CVs. | Open Source |
| 🛡️ | LLM Hallucination Detector | UQML-powered output uncertainty quantification platform evaluating whether LLM responses are factually grounded or hallucinated. | Open Source |
| 🏢 | Enterprise Functions AI Platform | Document intelligence & knowledge discovery engine for cross-department semantic search, workflow automation, and unstructured ingestion. | Enterprise |
| 📑 | AI MLR Compliance Assistant | Automated compliance validation engine using OCR extraction, transcript analysis, and regulatory knowledge retrieval to streamline review cycles. | Enterprise |
| 🚗 | Streamlit AI Parking Monitor | Computer vision application combining YOLOv8 vehicle tracking, EasyOCR license plate recognition, virtual zones, and SQLite telemetry. | Open Source |
| 📜 | Policy Document Analyzer | Conversational document intelligence platform enabling domain experts to ingest multi-page policies and execute contextual Q&A with citations. | Open Source |
| 🧪 | AI QA Automation Suite | Generative test scenario & script synthesis engine transforming Jira user stories into edge-case test matrices and automation scripts. | Enterprise |
📊 Click to expand interactive Mermaid architecture pipeline
flowchart TB
subgraph S1 ["1. Ingestion & Preprocessing"]
A["📄 Multi-Source Ingestion<br/><i>PDFs • Databases • Streams</i>"] --> B["⚡ Chunking & OCR Extraction<br/><i>YOLOv8 • EasyOCR</i>"]
end
subgraph S2 ["2. Dual-Tier Hybrid Retrieval"]
B --> C1["🧠 Dense Vector Search<br/><i>FAISS • PgVector • Pinecone</i>"]
B --> C2["🔍 Sparse Lexical Search<br/><i>BM25 Keyword Index</i>"]
C1 & C2 --> D["🎯 Reciprocal Rank Fusion & Cross-Encoder Rerank"]
end
subgraph S3 ["3. Agentic Orchestration & Tools"]
D --> E1["🤖 Primary Coordinator Agent<br/><i>Goal Decomposition & Planning</i>"]
E1 <--> E2["🛠️ Dynamic Tool Calling<br/><i>APIs • Code Exec • Web Crawl</i>"]
end
subgraph S4 ["4. Safety Guardrails & Delivery"]
E1 --> F{"🛡️ Grounding & Guardrails<br/><i>UQML Hallucination Check</i>"}
F -- "⚠️ Hallucination Detected (Reflection)" --> E1
F -- "✅ Validated & Grounded" --> G["🚀 High-Throughput Production API<br/><i>FastAPI Async • Low-Latency Streaming</i>"]
end
style A fill:#0e1526,stroke:#38bdf8,stroke-width:1.5px,color:#f8fafc
style B fill:#0e1526,stroke:#818cf8,stroke-width:1.5px,color:#f8fafc
style C1 fill:#0e1526,stroke:#a855f7,stroke-width:1.5px,color:#f8fafc
style C2 fill:#0e1526,stroke:#a855f7,stroke-width:1.5px,color:#f8fafc
style D fill:#0e1526,stroke:#ec4899,stroke-width:1.5px,color:#f8fafc
style E1 fill:#1e1b4b,stroke:#6366f1,stroke-width:1.5px,color:#e0e7ff
style E2 fill:#0e1526,stroke:#ec4899,stroke-width:1.5px,color:#f8fafc
style F fill:#2e1065,stroke:#f59e0b,stroke-width:2px,color:#fef08a
style G fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#d1fae5
- Deterministic Guardrails: Zero tolerance for ungrounded hallucination in enterprise setups; applying semantic confidence scores and structured schema enforcement.
- Hybrid Retrieval Over Naive Search: Pairing dense vector representations (FAISS/PgVector) with sparse BM25 indexing and cross-encoder reranking for pinpoint context precision.
- Asynchronous Scalability: Backends engineered with FastAPI async primitives, thread-pooled IO, and resilient connection pooling.
- Human-in-the-Loop Multi-Agent Systems: Empowering autonomous agents with clear tool definitions, self-correction loops, and deliberate approval checkpoints.
3D Contribution Graph
📈 Click to view detailed language & profile telemetry
I am always open to discussing cutting-edge Generative AI architectures, Agentic workflows, enterprise RAG consulting, or full-time opportunities.
Bandaragama, Sri Lanka | AI Engineer & Generative AI Architect | 1st Place Champion, Virtusa Hackathon
Built with precision, passion, and engineering rigor • © Lasitha Dilshan Thilakarathna