Senior engineer, 14+ years in distributed systems. I build anti-financial-crime systems at a Nordic bank by day, and solo-ship agentic-AI products — and win hackathons — by night.
I care about AI that explains its reasoning, leaves an audit trail, and actually works in production.
- 🏦 Software Engineer @ Swedbank — anti-financial crime & AML
- 🤖 Building governed agent fleets on Azure AI Foundry, Google ADK, A2A and MCP
- 🧮 Researching GPU & quantum compute — q1729, the quantum taxicab
🛡️ ARGUSMulti-agent compliance intelligence Microsoft Agents League — AI Skills Fest 2026 · Reasoning Agents track ↗
problem> Manual KYC/AML review doesn't scale, and unaudited AI decisions don't survive a regulator's audit. approach> Five specialist agents coordinated over A2A on Azure AI Foundry. Findings are cited via Foundry IQ, with an audit trail designed for regulatory scrutiny. impact> Citation-grounded regulatory lookups with reproducible audit trails.
|
🏰 BASTIONA governed institutional-agent fleet for continuous access review All Things Agentic Hackathon 2026 · Fortified Enterprise Fleet track ↗
problem> Access review is quarterly work performed on continuously changing permissions. Automating the scan isn't enough — an institutional agent must audit the real policy rather than fixture rows, survive asynchronous retries, prove why it acted, and remain unable to turn suspicious input into a privileged write. approach> Read-only IAM review against the GCP project that runs it, including its own service identities. Deterministic code detects and scores findings; Gemini explains and routes already-minimized risk. Three institutional agents, one durable investigation identity — no raw IAM binding crosses the model or human-notification boundary. impact> Human review receives minimized risk categories rather than raw IAM bindings.
Case study → · Code ↗ · Live app ↗ · Demo video ↗ · Devpost ↗ · Write-up ↗ |
Durable incident memory for cold-started agents CockroachDB × AWS Hackathon 2026 — Build with Agentic Memory ↗
problem> The conditions that cause production incidents — resource exhaustion, node failure, rollbacks, autoscaling churn — are the same conditions that kill the agent responding to them. An agent holding its working state in process memory does not degrade when that happens: it stops, and a human restarts the incident from zero without knowing which remediation actions already ran. approach> Every state transition is committed to CockroachDB before and after it happens. Kill the process mid-step — no graceful shutdown, no checkpoint call — and the next cold invocation reads the durable state, sees a step frozen in impact> Resumes an interrupted incident from the exact step it was killed on.
Case study → · Code ↗ · Live app ↗ · Demo video ↗ · Devpost ↗ |
-FFB000?style=flat-square){kind=icon}
Archived — hosting retired; the source, recordings and submissions remain.
- 📡 DRIFT — GPU & AI infrastructure release intelligence. case study · code · demo video · devpost
- 🎁 BANKERS' WRAPPED — A narrated financial recap video, generated end to end. case study · code · demo video · devpost
-
Agents League — celebrating the builders — techcommunity.microsoft.com · Microsoft's winners announcement, ARGUS named 1 of 3 for Hack for Good
-
ARGUS: Compliance Infrastructure That Believes Financial Access Is a Human Right — techcommunity.microsoft.com · Guest post
Microsoft published my full write-up on the Educator Developer Blog, including how ARGUS fans four specialist agents out over A2A and fans their findings in to a fifth, with citation-grounded risk scoring.
Lee Stott · Microsoft in #agentsleague (theme-aware image)
🧩 Agentic AI & Enterprise Intelligence I design AI systems with Azure AI Foundry, Foundry IQ, multi-agent orchestration, Agent-to-Agent (A2A) communication, and RAG with hybrid search — built to be explainable, grounded, and production-ready.
🏗️ Distributed Systems & Backend Architecture 13+ years building scalable platforms with Java (Spring Boot, Quarkus) and Python (FastAPI) — microservices, NoSQL, event-driven systems, and hybrid cloud across AWS, Azure, and OpenShift.
⚙️ AI Infrastructure, Performance & Compute I work at the infrastructure layer behind modern AI — GPU computing, NVIDIA CUDA, model serving, vector search, and performance engineering.
📡 Reliability, Observability & Platform Engineering I build resilient systems with Azure Functions, Service Bus, OpenTelemetry, Application Insights, and KQL — telemetry pipelines that hold up to enterprise-grade reliability and governance.
Languages & frameworks
Agentic AI & LLM
Cloud & infrastructure
AI infra, GPU & observability
2025 – Present · Software Engineer · Swedbank — Stockholm, Sweden Anti-financial crime · AML platforms · 95%+ test coverage across unified multi-module architecture
2021 – 2025 · Senior Java Developer · Viaplay Group — Stockholm, Sweden Media & streaming on AWS + Kubernetes · ~30% perf gains · ~40% delivery-speed acceleration
Mar–Sep 2021 · Software Developer · Expleo Technology Nordic — Gothenburg, Sweden Domain-driven microservices · ~50% faster onboarding via docs & workflow diagrams
2012 – 2021 · Senior Software Engineer · IBM — Sydney & Pune Regulated banking APIs for Westpac · ~25% response-time gains · Jenkins/Bamboo modernization
- q1729 — On an RTX 5070 Laptop GPU, the CUDA implementation reached double-precision saturation (~16 digits) in ~2.6 ms, while simulated QAE reached ~5 digits in ~0.44 s. No crossover was observed on this hardware.
- llm-qlab — Across three 7B model families, decode throughput fell monotonically as quantized weight size increased. Full Llama-2 offload measured 6.2x CPU-only decode throughput.
- pythonic-algorithms-lab — In the published sweep, radix sort is the strongest GPU result, while BFS and reductions remain slower than their CPU baselines at the tested sizes.
- iq-series — Completed Microsoft IQ learning cookbooks with executed Foundry IQ notebook outputs.
- NVIDIA Certified Professional: Agentic AI — NVIDIA
- Kubernetes Administration — The Linux Foundation
- IBM Machine Learning Essentials — IBM
Full certification list on LinkedIn ↗
Building trustworthy AI systems that explain their reasoning, leave an audit trail, and actually work in production.
If that's the kind of problem you're working on — I'd love to talk.
github.com/iarjunganesh · arjunganesh.dev