AI & LLM Systems
Where I spend most of my time retrieval, orchestration, and getting models to production.
- Transformers
- RAG
- Agentic AI
- LangChain
- LangGraph
- Fine-tuning
- Unsloth
- Inference Optimization
- Autoencoders
- LSTMs
Open to AI/LLM engineering roles
Kavishka Kalhara
Computer EngineerLLM Systems Engineer, AI Security Researcher, Full-Stack AI Developer
I build secure, production-grade AI systems from RAG pipelines and agentic orchestration to self-healing security architectures that detect, explain, and repair their own failures.
Matara, Sri Lanka
Currently
BEng Computer Engineering
University of Ruhuna
Focus
LLM Security
About
I build AI systems that hold up outside the demo.
I'm a Computer Engineering undergraduate at the University of Ruhuna, working at the intersection of LLM systems, scalable AI backends, and secure distributed architecture. Most of what I enjoy sits in the unglamorous middle: retrieval that actually returns the right chunk, inference that stays fast under load, and services that fail safely instead of loudly.
I spent six months as a software engineering intern shipping production AI applications RAG pipelines, prompt orchestration, and model-serving infrastructure that real users depended on. Alongside that I've run IoT security research, and founded my faculty's Computer Engineering Society from an empty room and a proposal document.
Right now my focus is AI safety and defensive engineering: applying the OWASP Top 10 for LLMs to agentic systems, and designing architectures that detect, explain, and recover from their own failures.
6 mo
Production internship
Shipping real AI features
3+
Major systems built
Security, alumni, LLM infra
6+
Leadership roles
Societies, exhibitions, CTFs
2
National competitions
1st Runner-Up · Finalist
Toolkit
The stack I reach for.
Grouped by where it sits in the system from model internals down to the metal it runs on.
Languages
- Python
- TypeScript
- JavaScript
- C / C++
- Java
Backend
- Node.js
- Express.js
- FastAPI
- REST APIs
Cybersecurity
Defensive engineering threat modelling LLM apps before they ship.
- Web Application Security
- OWASP
- LLM Vulnerability Mitigation
DevOps & Cloud
- Docker
- Kubernetes
- AWS EC2
- Nginx
- Linux
- ArgoCD
- Jenkins
Databases
- MongoDB
- PostgreSQL
- SQL
- ChromaDB
Frontend
- React.js
- Tailwind CSS
Security Tooling
- Wireshark
- Nmap
- Kali Linux
Parallel Computing
- CUDA
- OpenMP
- MPI
AI Coding Tools
- Claude Code
- GitHub Copilot
Experience
Where I've shipped.
Production work the kind with users, deadlines, and on-call consequences.
Intern Software Engineer
BotCalm (Pvt) Ltd
May 2025 — Nov 2025Sri Lanka
Six months building LLM products that had to survive real traffic, real users, and real latency budgets.
- Built production LLM applications on top of RAG pipelines semantic search, prompt orchestration, and vector-based knowledge retrieval serving live use cases.
- Designed the secure AI backend around them: API authentication, async job processing, and model-serving infrastructure that scaled without falling over.
- Wired multi-step reasoning pipelines with n8n and LangChain, connecting external tools into automation flows the team could trigger and trust.
- Pushed retrieval quality up through better embedding strategies, context filtering and prompt refinement then chased the latency back down with GPU performance tuning.
Stack
- Python
- LangChain
- Unsloth
- Hugging Face
- RunPod
- Docker
- n8n
- React
- Tailwind CSS
BEng (Hons) Computer Engineering
Faculty of Engineering, University of Ruhuna — undergraduate
Selected Work
Systems I've designed and shipped.
Three builds that cover the range: applied security research, a production platform, and an architecture experiment.
3 projects
View project
2025Research · AI Security
AI-Driven Self-Healing Intelligent System
IoT security that explains itself — and repairs itself.
An IoT security framework that watches network traffic, detects anomalies with sequential ML, explains its reasoning, and lets LLM agents plan and execute the fix.
- Kafka
- FastAPI
- Zeek
- GRU
- SHAP
- Multi-Agent
View project
2025Full-Stack · Production
refaa.lk — Faculty Alumni Platform
A production platform for the Ruhuna Engineering alumni network.
A secure alumni platform with hardened auth, live payments and automated CI/CD — tuned with Nginx load balancing, Redis caching and Kafka events to stay fast under concurrency.
- React
- Node.js
- MongoDB
- Redis
- Kafka
- AWS
View project
2025AI Systems · Research
AI Chatbot — Dual-Model Continuous Learning
Serve and learn at the same time, without dropping a request.
An Active-Shadow architecture that answers in real time while a shadow model fine-tunes in the background, then promotes atomically via weight symmetry.
- Python
- Unsloth
- LangGraph
- DeepSeek-R1
- FastAPI
Recognition
Competitions won, communities built.
Two things I care about equally: solving hard problems under a clock, and leaving structures behind that outlast me.
Achievements
1st Runner-Up
BitCode v5 — Competitive Programming Competition
University of Ruhuna
National Finalist
Robofest 2024 — National Robotics Competition
Sri Lanka
Leadership & volunteering
Founding President
Computer Engineering Society (ComES)
University of Ruhuna
Founded the society from proposal to active chapter.
Main Organizer
ReXtro 2025
National Engineering Exhibition
Co-Chairman
HaXtreme 3.0 & Red Cypher CTF
Capture-the-flag competition
Main Organizer
IEEE Student Branch
University of Ruhuna
Program Lead
CareerXpo 1.0
Careers & industry expo
Program Lead
HaXtreme 2.0
Competitive Programming Competition
Team Lead — Brand Marketing
AIESEC in Ruhuna
Youth leadership organisation
Writing
Notes from the build
I write about LLM systems, retrieval design and the security side of shipping AI — mostly the things I wish someone had written down before I hit them.
Contact
Let's build something that ships.
I'm open to AI/LLM engineering roles, research collaborations, and interesting problems in secure systems. Drop a message and I'll get back within a day or two.
Open to opportunities · 2026 intake
Reach me directly
- Emailkavishkakalharapro@gmail.com
- PhoneAvailable on request
- LocationMatara, Sri Lanka