Arjun GaneshAgentic AI · Distributed systems
02

Selected systems

Five systems in production or active development. Problem, approach, and what each one actually guarantees.

Multi-agent compliance intelligence

Winner · Microsoft Agents League 2026 · Hack for Good (1 of 3)

Problem

Manual KYC/AML review doesn't scale, and unaudited AI decisions don't survive a regulator's audit.

Solution

Five specialist agents coordinated over A2A on Azure AI Foundry. Every finding is cited via Foundry IQ. Full audit trail. All decisions are explainable, reproducible, and regulatory-proof.

Impact

100% auditable, citation-grounded regulatory lookups replacing manual screenings.

A governed institutional-agent fleet for continuous access review

Problem

Access review is quarterly work performed on continuously changing permissions. Automating the scan isn't enough — an institutional agent must remember prior human decisions, survive asynchronous retries, prove why it acted, and remain unable to turn suspicious input into a privileged write.

Solution

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

Humans receive counts and allowlisted categories, never bindings. 161 tests at 100% statement and branch coverage.

GPU & AI infrastructure release intelligence

Problem

Raw changelogs are noisy, unstructured, and full of false positives. Teams miss critical AI infrastructure updates.

Solution

High-precision release aggregation. Raw data → dependency checks → bounded, technical summaries. Built with FastAPI + pgvector semantic deduplication.

Impact

Converts raw, noisy changelogs into actionable release intelligence.

dr1ftless.vercel.appopen ↗

Durable incident memory for cold-started agents

Problem

Cold-started agents lose execution state. Multi-step workflows restart from zero, wasting compute and losing context.

Solution

Distributed checkpoint engine on CockroachDB. Restores 100% of execution state without pipeline restart. Mission-critical runtime guarantees.

Impact

Restores full execution state across distributed orchestration.

huggingface.co/spaces/iarjunganesh/continuumopen ↗

Year-in-review intelligence for financial workflows

Problem

Data-centric fintech teams want year-in-review insights. Existing tools are generic, not built for financial workflows.

Solution

Spotify Wrapped but for banking. Extracts transaction intelligence, generates insights, creates shareable year-end summaries.

bankers-wrapped.arjunganesh.devopen ↗