Two ideas in production

LMA

01 The System

未来 Mirai

The operating system for audit preparation.

A three‑layer engine for verifiable financial work. Every number on every page is computed and verified in code before any model sees it. Every conclusion is a typed claim with evidence references and rule lineage. Every claim carries a confidence tier bound to its weakest dimension — so the operator only sees what actually needs human judgment.

Built for a lane the largest accounting firms structurally cannot serve. SEC independence rules block the Big Four from preparing the financial statements they later audit, leaving private companies, first‑time auditees, and pre‑IPO issuers to specialty boutiques that price by the hour and cannot scale. Mirai productizes the work — quarterly conversion, pre‑audit conversion, IPO‑grade restatement — at fixed engagement tiers with a deliverable an auditor can replay end to end.

01 Deterministic substrate Arithmetic ties, balance‑sheet equation, cross‑period reconciliation — tolerance: $0. The class of LLM arithmetic error is eliminated before any model is invoked.
02 Assertion graph Every conclusion is a typed claim with evidence, rule lineage (ASC sections, skill files), sub‑assertions, and approval history. Built for inspection, not just inspection‑ready.
03 Five‑dimension confidence Extraction, Mechanical, Classification, Judgment, Completeness. Each assertion's tier is bound by its weakest dimension — review compresses to the moments that actually require human judgment.
02 The Strategy

SkyProp

An internal multi‑asset trading strategy.

NoticeSkyProp is an internal vehicle operated for the principal's own account. It is not a registered fund, is not offered to outside investors, and does not accept subscriptions of any kind.

SkyProp combines two synergistic strategies — dynamic multi‑asset allocation and opportunistic energy trading — automated end to end through proprietary AI infrastructure. Built on methods developed at Higher Moment Capital, ranked the number‑one CTA over five years by BarclayHedge.

The internal thesis: institutional‑grade active management has historically required institutional‑grade operating cost. Automating the full value stack — research, signal generation, portfolio construction, execution, and compliance monitoring — collapses that cost structure for a single‑account vehicle.

01 Dynamic asset allocation Proprietary quantitative process across equities, fixed income, commodities, currencies, and real estate. Mean‑expected‑shortfall optimization with daily parameter updating, rebalancing, and risk‑targeted construction.
02 Energy trading Opportunistic long and short positions in energy futures, capturing seasonal volatility, supply‑demand dislocations, and term‑structure dynamics through quantitative signal‑based entries and exits.
03 AI automation Agentic infrastructure runs research, portfolio construction, execution, and 24/7 risk and compliance monitoring — delivering institutional‑quality operations at a fraction of traditional fund operating cost.
03 About

About LMA

LMA is the personal investment and operating vehicle of Sundar Subramaniam. Based in Cambridge, Massachusetts. Background and prior work at ifesia.com.