# Aditya Dalmia Staff-level backend engineer and builder with about 7.5 years of experience across distributed systems, AI-agent and reinforcement-learning evaluation, product architecture, and engineering leadership. Has built high-throughput services, evaluation environments, and products from early design through delivery. ## Canonical profiles - Human site: https://adityadalmia.com/ - Professional profile: https://adityadalmia.com/about.md - Résumé: https://adityadalmia.com/resume.md - JSON Resume: https://adityadalmia.com/resume.json - Project data: https://adityadalmia.com/projects.json - Writing index: https://adityadalmia.com/blog/ - RSS: https://adityadalmia.com/rss.xml ## Projects - Helmsman: https://adityadalmia.com/projects/helmsman/ — Helmsman brings together research engines, shared data infrastructure, an operator workspace, and Structure Lab. Built independently, with a record of what we tried and what we learned. Status: Private research platform. - RL environments: https://adityadalmia.com/projects/rl-environments/ — Built reinforcement learning environments, the agents that generate them, and benchmarks for difficult engineering work at Proximal Labs. Contributor to FrontierSWE. Status: Work at Proximal Labs, February–June 2026. - Rental OS: https://adityadalmia.com/projects/rental-os/ — An open-source prototype of rental operations across tenants, owners, operations staff, guards and admins. It explores shared identity, backend authorization and the handoffs between visits, bookings and agreements. Status: Open-source prototype. - BNC Cloud: https://adityadalmia.com/projects/bnc-cloud/ — A Kubernetes platform I designed and run on Hetzner: cloud workers, a dedicated server for memory-heavy work, and a burst pool that scales to zero. More than forty components sit behind Git-driven deployments. This site runs on it. Status: Independent lab platform. - Cross-repo Skill Registry: https://adityadalmia.com/projects/skill-registry/ — A shared knowledge and skill platform across repositories and machines. MCP discovery, hybrid search, provenance, verification, and explicit relationships between entries. Status: Private agent infrastructure. - Flightdeck Tools: https://adityadalmia.com/projects/agent-tooling/ — An engineering case study in coordinating work, preserving state, and checking evidence. Eleven projects with thirteen tools, published as public Apache-2.0 reference implementations. Status: Deprecated for new Claude Code integrations since September 22, 2026. ## Published writing (Markdown) - Helmsman: when a research candidate becomes infrastructure: https://adityadalmia.com/blog/helmsman-promotion-boundary.md — The design of an explicit promotion boundary: frozen definitions, concurrent requests, durable evidence gates, and the limits of a deployed research workspace. - Rental company management is a handoff problem: https://adityadalmia.com/blog/rental-os-company-workflow.md — Following a property from a guard's lead through visits, negotiation, documents and closure—and the state machines, checks and boundaries each handoff needs. - Inside the deal room: turning conversation into rental terms: https://adityadalmia.com/blog/rental-os-inside-the-deal-room.md — How Rental OS separates private discussion, versioned proposals, bilateral sign-off and closure evidence—and how one missing state-machine edge broke every signed counter-proposal until it was fixed. - Where the boundaries sit in Rental OS: https://adityadalmia.com/blog/rental-os-system-boundaries.md — How Rental OS connects five portals to shared records, authenticated reactive queries and asynchronous work, how every public Convex function was swept for authorization before the source went public, and which integrations remain unfinished. - Evidence without automatic green lights: https://adityadalmia.com/blog/agent-tooling-evidence-without-green-lights.md — Why repository health, configuration baselines, and review receipts distinguish a verified observation from approval or certainty. - Extracting the useful parts of a private agent rig: https://adityadalmia.com/blog/agent-tooling-extracting-the-useful-parts.md — Why I split a large, machine-shaped agent system into small tools with explicit contracts, synthetic examples, and new histories. - Guarding the dangerous moment before a local change: https://adityadalmia.com/blog/agent-tooling-guarding-local-change.md — How gates, locks, Git preflights, and file guards reduce ambiguity around checks, worktrees, and agent-written files. - State that survives the terminal: https://adityadalmia.com/blog/agent-tooling-state-that-survives-the-terminal.md — Four small tools for durable messages, lifecycle history, generation-safe progress, and decisions whose original wording remains intact. - Building the cloud behind my projects: https://adityadalmia.com/blog/cloud-building-my-own-platform.md — A small Hetzner Kubernetes platform with shared services, repeatable deployments, and explicit operational tradeoffs. - When every small app brings its own connection pool: https://adityadalmia.com/blog/cloud-postgres-pools-and-proof.md — The shared Postgres limit, the rollout that reduced idle connections, and why a green dashboard is only one kind of evidence. - Self-hosting Temporal and putting it to work: https://adityadalmia.com/blog/cloud-self-hosted-temporal.md — The infrastructure integration and application workflows behind our self-hosted Temporal deployment. - Two network planes, two different responsibilities: https://adityadalmia.com/blog/cloud-two-network-planes.md — Why WireGuard protects cluster administration while NetBird connects applications and agent tools. - Building Helmsman as a platform: https://adityadalmia.com/blog/helmsman-building-a-platform.md — How one research engine grew into a shared data layer, an operator workspace, and a place to test competing ideas. - Helmsman: from research candidate to monitor: https://adityadalmia.com/blog/helmsman-control-plane.md — Stable identity, evidence gates, and visible deployment states in the Helmsman operator workspace. - A data architecture that remembers what it knew: https://adityadalmia.com/blog/helmsman-data-architecture.md — One synthetic corrected value traced through the real observation writer and schema: revisions as new rows, an as-known-at query, frozen run inputs, and freshness judged against schedule and consumer. - What we kept from the experiments we set aside: https://adityadalmia.com/blog/helmsman-experiments-we-left-behind.md — Polymarket research, unreliable causal graphs, and the value of preserving a rejection with its evidence. - Moving the scheduler without running the work twice: https://adityadalmia.com/blog/helmsman-moving-schedulers.md — A cutover trace from Helmsman's move to Kubernetes CronJobs: one declared owner per job, a revision-pinned switch, a rollback that could not be a revert, and alerts that had to move too. - Building a real-time news parsing pipeline: https://adityadalmia.com/blog/helmsman-realtime-news-parsing.md — One synthetic headline traced through a partial observation, an upgrade, duplicate deliveries, an edit and a retraction, using the ingestion writer's real storage contract. - Score once, experiment many times: https://adityadalmia.com/blog/helmsman-score-once-experiment-many.md — A synthetic identity record traced through the real score store: which change needs new model work, which only reruns deterministic math, and why replaying stored scores does not recover the past. - Turning a chart argument into a testable definition: https://adityadalmia.com/blog/helmsman-structure-lab.md — One vague chart claim turned into three versioned definitions and replayed on synthetic data through Structure Lab's real detector and registry. The replay shows what a new version does to earlier annotations and evidence. - Where the model stops and the code takes over: https://adityadalmia.com/blog/helmsman-where-models-stop.md — The contract between model output and deterministic code in Helmsman, traced with synthetic replies: what a model may emit, what the validator rejects, and what happens when the handoff fails. - One person, many roles: designing Rental OS: https://adityadalmia.com/blog/rental-os-one-person-many-roles.md — How a rental-company prototype connects customers, owners, field teams, guards, and administrators without treating a role switch as permission—now open source under MIT. - An RL environment is a systems problem: https://adityadalmia.com/blog/rl-environments-are-systems.md — A task, a sandbox, and a score need to agree. The engineering around that contract is where things get interesting. - A shared memory for agents across repositories: https://adityadalmia.com/blog/skill-registry-cross-repo-knowledge.md — Inside the skill registry: discovery, provenance, verification, and sharing knowledge without copying it into every repo. ## Reading notes Only published work appears here. Private source repositories, draft articles, internal evidence notes, and operational records are intentionally excluded. Lifecycle and evidence dates describe what was verified; they are not promises of current public access.