Rishi Kanajam

Full-stack & AI engineer in Sydney

I build AI systems that know what they don't know.

Full Australian work rights through Sept 2028.

Selected work

Shipped

Krama Core

Open-source FHIR R4 and compliance SDK for India's ABDM health network. Published to PyPI, Apache 2.0 licensed.

One decision: Compliance returns blockers, not warnings.

SourcePyPI
Evidence
v1.0.0-alpha.3 on PyPI
128 tests; CI on Python 3.10–3.12
Apache 2.0
Limits
Australia, UK and US adapters: metadata only, no live exchange
Built with
Python · FHIR R4 · Pydantic · Cryptography

Shipped

PathFinder

AI triage and referral platform for the Lake Macquarie & Newcastle Suicide Prevention Network. Built with a teammate at NGM Group's hackathon, May 2026.

One decision: Crisis numbers survive the backend.

Source
Evidence
Idea Disruptor Award, NGM Group hackathon
35+ documented edge cases
Built with a teammate
Built with
FastAPI · Azure AI Language · React · Python

In progress

DeceptionArena

Cross-game LLM deception benchmark: five social deduction games behind one engine interface, with structured decision traces. What exists is benchmark infrastructure, not findings.

One decision: Score the public rationale, not the chain-of-thought.

Source not public yet
Evidence
1000/1000 randomised stress tests
15 models across 5 providers wired in
Limits
Real-API runs so far are small
Infrastructure, not findings
Built with
Python · asyncio · aiohttp · SQLite

Shipped

ChronoLens

Cultural research workspace that labels every AI claim by how far you should trust it. Built solo in about five hours at the OpenAI Codex Hackathon, Sydney.

One decision: Every claim carries its epistemic status.

Source
Evidence
Built solo in about five hours
Ten analysis surfaces, six live archive APIs
Built with
Next.js 14 · TypeScript · GPT-4.1 · Multi-API

More work

  • PDF-RagHybrid RAG: BM25 + dense retrieval with RRF, cross-encoder re-ranking, and an LLM-as-judge evaluation harness.
  • Pinch Recovery EngineFailed-payment recovery engine for Pinch Payments: per-dishonour-code strategies, AU business-day scheduling, reasoned decisions.
  • MedVaultMulti-tenant clinical AI platform: Gemini decision support with physician review, IoT cold-chain, offline-first records.
  • MangRAGRAG service over PDFs: MongoDB Atlas Vector Search, local embeddings, Groq LLaMA 3.3, served via FastAPI, Streamlit and CLI.
  • anomaly-detectorLog anomaly detection with Isolation Forest, One-Class SVM and an ensemble vote; Flask UI, Dockerised.
  • DSA_Learning_WebInteractive platform for learning data structures: write code and watch the structure change step by step.

More on GitHub →

Experience

  • Apr 2026 – PresentEngineer & Maintainer, Krama CoreNirvya Labs
  • Jan 2024 – PresentAcademic Tutor, Data Structures & Algorithms (COMP2123/COMP9123)University of Sydney
  • Apr 2025 – Mar 2026Systems Engineer – Data Team (Part-time)St Vincent de Paul Society NSW
  • Aug 2024 – Dec 2024Computer Vision Engineer (University Capstone)Ocius Technology
  • Mar 2023 – Jul 2023Frontend Developer (Contract)HRVITE Services Pvt Ltd
  • Jun 2022 – Jan 2023Software Engineer (Trainee)Revidd

About

Every system I've built has the same signature: it tells you how much to trust it. I build AI systems that know what they don't know.

I'm from Andhra Pradesh (Telugu). When my grandmother needed care, our records travelled on paper. For many underserved communities in India, the nearest hospital can be 60 kilometres away; they're who this work is for. That's why I maintain Krama Core, an open-source FHIR and compliance SDK for India's ABDM health network, published under Nirvya Labs.

Background, experience and education →