Work & Experiments
We demonstrate real capability before claiming scale. Each case study below reflects production software, multi-agent systems, or foundational language research shipped by our team.
Molarity: AI-Powered Clinical Simulation & Assessment
Molarity explores how conversational AI can simulate authentic clinical examination scenarios and provide structured, rubric-aligned candidate assessment and evidence-linked feedback. Rather than relying solely on static flashcards or multiple-choice questions, candidates engage in natural voice consultations with simulated patient personas.
Technical Architecture Highlights
Dynamic prompt state machines acting out symptomatic illness, emotional hesitation, and history revelation.
High-throughput audio streaming enabling human-speed medical dialogue directly in the browser.
Multi-dimensional evaluation measuring history-taking thoroughness, empathy, differential diagnosis, and timing.
Structured remediation reports providing timestamped dialogue evidence for candidate revision.
✓ Associated autonomic features: Nausea/Diaphoresis (Passed)
Score: 94% on History Thoroughness
SiBuzz: Multi-Agent Sinhala Media Publishing System
SiBuzz was built as an end-to-end experiment in autonomous multi-agent content research, natural language synthesis, and automated publishing tailored for Sinhala-speaking diaspora communities in the United Kingdom.
Rather than viewing it as a conventional media outlet, Susinidu treated it as an architectural test: can an agentic loop reliably ingest real-time news, synthesize culturally coherent Sinhala articles, verify factual consistency, and publish autonomously to production without human-in-the-loop bottlenecks?
Demonstrated that multi-agent research and localized translation agents can operate reliably at scale. Served actual traffic to UK readers with high uptime, validating our pipeline architecture.
Sinhala TTS: Neural Speech Synthesis Research
One of the earliest foundational research efforts at Susinidu Deep Tech. While major global models prioritized English, Spanish, and Mandarin, Sinhala remained underrepresented in natural-sounding neural acoustic text-to-speech.
We engineered custom phonetic mapping systems, tokenizers for complex Sinhala script conjuncts, and trained lightweight acoustic models capable of synthesizing conversational speech for voice assistants and automated reading.
This speech technology laid the foundation for our ongoing work in low-latency voice pipelines, which today powers interactive clinical actors in Molarity.
Interactive Web 3D & Procedural Simulation
Currently engineering lightweight WebGL and WebGPU interactive toys, browser 3D physics playgrounds, and realtime multiplayer character sync.
Explore the interactive labs prototypes