Alexander Derkatsch

About

Aerospace Systems Engineer → AI Orchestrator. 3+ years on NASA SLS Block 1B (Artemis 4+) at Beyond Gravity — requirements derivation, V&V, DOORS hazard tracing, NASA / MIL / DoD compliance, sole-site IPT lead — paired with a deep portfolio of production AI / LLM systems shipped across propulsion mission-control simulation, classical GNC, LLM-driven legacy-system integration (with air-gapped classified routing), multi-agent training environments, and AI-augmented manufacturing intelligence. Bridge between legacy aerospace rigor and the AI-augmented engineering loop now reshaping the propulsion and missile-systems pipeline.

The Bridge: Legacy Aerospace Rigor + AI-Augmented Engineering Loop

Three years inside NASA human-rating taught me how safety-critical hardware actually gets to flight: requirements integration in DOORS, qualification-directed compliance assessments, hazard closure through verifiable test artifacts, IPT coordination across primes and international facilities, V&V plan development, and CFRP production for launch hardware. That work doesn't go away under AI — it gets multiplied. The atomic tasks (analysis runs, document synthesis, schema discovery, connector wiring) become machine-cheap; the valuable work is composition, judgment, and validation against domain constraints that matter.

What I've spent the last 18 months doing is the second half of that bridge: building production AI / LLM systems end-to-end as the orchestrator — multi-agent simulation, agentic codegen, RAG over million-token corpora, real-time voice agents, AST-policy sandboxing, air-gapped local-LLM routing for classified data, and CJIS-prep security postures. Not as a hobbyist; as a systems engineer applying the same rigor to AI architecture that I applied to flight hardware.

Current Work: bigBespoke LLC — One-Person Applied-AI Lab

Founded full-time post-Beyond Gravity (nights & weekends before that), bigBespoke is a portfolio of production systems each chosen to teach a different facet of the AI-augmented engineering loop:

T-MINUS — Real-time multi-agent mission-control simulator orchestrating 8 LLM-driven console operators under a Flight Director command loop. G-FOLD lossless-convex landing guidance, CHUG / L1 / T1 combustion-instability modes, JONSWAP irregular sea-state droneship dynamics, emergent IMU / GPS / RCS / radar-altimeter part-failure cascades. Validated through an 8-flight Monte Carlo campaign with customer-specific operator personalities. The principle: "the monitor decides WHEN, the LLM decides WHAT and HOW."
FletchGNC — End-to-end classical GNC stack and a self-authored 44-chapter technical book, Beyond Proportional Navigation, taking the reader from the geometry of pursuit through symplectic, information-geometric, octonionic, and Koopman-spectral guidance. 9-state EKF at 1 kHz, plant-inversion control, RK4 1 ms physics, tactical-grade MEMS IMU model — production-quality reference architecture for a missile-seeker / orbital-rocket successor.
Universal Interface — LLM-driven middleware purpose-built for federal legacy-system friction (80% of federal IT spend goes to legacy maintenance; 53% of agencies still do manual data transfer). Discovers heterogeneous I/O, generates a Python connector, auto-synthesizes pytest cases, iteratively refines until tests converge. Pluggable provider router auto-selects between Grok 4.3 cloud and Qwen3 via Ollama for classified / restricted data — no manual switching. AST-policy + subprocess sandboxing on every generated connector.
CopApp.ai — Multi-agent LLM training platform for police de-escalation. Three-agent pipeline (Dialog → World → Psyche), automated ICAT-aligned grading, skill-tree progression, certification tiers. 60 commits / 14,000+ lines of TypeScript / 54 cloud functions / 22 routes. CJIS-prep posture from the rules layer up.
PawTrek — "The K9 Handler's Unfair Advantage." Voice-first capture (Deepgram + Grok) generates court-admissible narratives aligned with Florida v. Harris admissibility expectations. CMEK on Firestore, App Check with real attestation providers, offline-first SQLite mirror with Firestore outbox + per-collection watermarks. Beta with a sworn handler in Monroe County, AL.

Plus a tail of smaller-but-meaningful systems: bigBespoke Geometry AI Platform (STL/STEP/IGES analysis + hydrodynamics + printability scoring), rocketProcedure (orbital sim with neuromorphic adaptive GNC), Sentinel (multi-source weather trading with Kelly sizing), ChatWithRob (1.9 M-token RAG over 8 years of sermons), Astro Pixie / star.report, anyBible (14-translation probabilistic consensus), Lattice GUI, DizKord, misalignment-detector — all documented on the portfolio page.

NASA Artemis Experience

Beyond Gravity (2022 – 2025): Systems engineer for SLS Block 1B Universal Stage Adapter — the composite structure connecting Orion to the rocket for Artemis 4+. Owned the systems-engineering lifecycle: requirements integration, V&V, qualification campaigns, compliance assessments, RIO / scrum cadence, and CFRP resin-system designs for flight-rated production. Sole-site Systems Engineer / IPT lead for U.S. operations and program-level deputy. Authored human-rated safety and risk assessments under NASA, Army, and U.S. Space Force (DoD) standards; traced hazards in IBM DOORS to verifiable test artifacts; supplied measurable safety requirements into SOWs / specifications; adjudicated change, waiver, and exemption requests for the payload-adapter clamp-band assembly. Deployed 12 weeks to Linköping, Sweden to resolve a NASA-standard compliance deadlock with the Leidos-Dynetics prime on the SLS Payload Separation System. Drove a manufacturing-philosophy shift from Swiss sequential processing to American linear-adjacent flow; built a rate-plan for a tiger-team alternate production line targeting a two-week takt time on Vulcan's aft-skirt heatshield.

Research, Publications & Mathematical Work

Published at AIAA SciTech 2022 (#2022-0994) and INCOSE Detroit 2021 (#129) on managing complexity in high-reliability organizations through quantifying communication and collaborative intelligence — frameworks built on SysML / MBSE / DOORS and natural-language programming theory, directly anticipating today's AI-augmented engineering loops. Authored the 44-chapter technical reference Beyond Proportional Navigation covering classical and optimal guidance, modern nonlinear control, terminal ballistics, and advanced topics through symplectic, information-geometric, octonionic, and Koopman-spectral guidance. Authored a working paper on AI-NPC police-communication patterns underpinning CopApp.ai. Currently completing Wheel-Laurent Algebra: A Rigorous Formalization — peer-review-grade work that contributes a total symbolic algebra distinguishing four types of zero with provenance carried through indeterminate forms, while explicitly refuting prior overclaims on Yang-Mills resolution.

Background

Cal Poly Aerospace Engineering (2014 – 2021): Innovated a mathematical model integrating high-reliability-organization principles into engineering teams during 2020 – 2022 undergraduate practical research (40+ hrs/wk); framework directly anticipated today's AI-augmented engineering loops. Senior capstone: ground systems / comms / C&DH lead on a rapid-prototype spacecraft sized for 5-pixel imaging of an Interstellar Object at 10 km/s flyby. Cal Poly Hyperloop 2018 – 2019: top-22 of ~130 international teams at the SpaceX Hawthorne Pod Competition; authored brake-system test manuals and ran 80 → 0 mph deceleration experiments.

Princeton Airport (2013 – 2014; 2016 – 2017): Flight Coordinator managing schedules across 60+ active student pilots and 100+ aircraft renters under the airport director. Private Pilot License, August 2017 — Princeton Flying School.

What I Offer

Aerospace systems-engineering rigor: Requirements derivation, allocation, and documentation; V&V plan development; DOORS-heavy hazard tracing; functional analysis and trade studies; risk and opportunity assessment; IPT coordination; compliance under NASA, MIL-STD, Army, Space Force, and human-rating standards; change / waiver / exemption disposition. SysML / MBSE / CAMEO familiar.

Propulsion, GNC & simulation depth: Solid-motor adapter integration, payload separation systems, human-rating, CFRP production. Classical GNC stacks (Proportional / Augmented PN, ZEM / ZEV optimal guidance, NDI control, sliding-mode, plant-inversion), G-FOLD lossless-convex landing guidance, combustion-instability modeling (CHUG / L1 / T1), 9-state EKF and nonlinear filtering, observability analysis, Monte Carlo trajectory and UQ, powered explicit guidance, actuator modeling, thrust vectoring, RCS allocation.

Applied AI / agentic systems: Multi-agent LLM orchestration (Claude, Grok, GPT) with monitor-driven event pipelines; air-gapped local-LLM routing for classified / restricted data (Qwen3 via Ollama) with cloud-LLM fallback for unclassified; agentic codegen (Claude Code, Cursor); LLM-driven schema discovery, connector generation, and iterative test-refinement; AST-policy sandboxing; real-time voice agents (Deepgram + xAI); RAG over million-token corpora; MCP / tool-use integration; Firebase / serverless production deployment; OIDC + PKCE, JWKS validation, token-bucket rate limiting, idempotency keys, CMEK, App Check, CJIS-prep audit logging.

Programming & ML: Python (heavy), MATLAB / Simulink, C / C++ (working), TypeScript / React / Node, PyTorch, NumPy, SciPy, OpenCV, FilterPy, Three.js. Comfortable working alone (building products end-to-end) or in large teams (NASA / prime contractor environments).

A Working Thesis

The Garden

The resume answers what I've done. The Mission answers what I'm doing it for — and why every project below is a wedge, not a destination.

Kibble → real food. Dogs first, then humans. Sourced inside a 60-mile radius and delivered milkman-style. Agrihoods where neighbors supply each other and children grow up in the garden — proverbially, figuratively, and literally. Bespoke houses tailored by AI conversation into a full CAD model with BOM, supply list, and visual work instructions — built by robots where they make sense and humans where they don't. Then planes, cars, boats, coffee, anything.

Rapidly-customizable simulations of every servant-served power dynamic — doctor-patient, teacher-student, engineer-customer — so the served party walks in prepared. Mars is the limit case; the garden is the daily reality. The human is the core for empowering. The healing is the heart. The proverbial garden is the goal.

Read the full thesis

Featured Projects

Five production AI / LLM systems shipped through bigBespoke LLC. Each was chosen to teach a different facet of the AI-augmented engineering loop — from multi-agent simulation to air-gapped local-LLM routing. Click any card for the full write-up.

Multi-Agent Simulation Validated

T-MINUS: Propulsive Landing Protocol

Real-time multi-agent mission-control simulator for propulsive landing

Orchestrates 8 LLM-driven console operators (Propulsion, GNC, Structures, Weather, Thermal, Recovery, Inspection, ASDS) under a Flight Director command loop (HOLD / RESUME / SCRUB / GO-NO-GO). Implements G-FOLD lossless-convex landing guidance, CHUG / L1 / T1 combustion-instability modes, JONSWAP irregular sea-state droneship dynamics, and emergent IMU / GPS / RCS / radar-altimeter part-failure cascades.

Core principle: "the monitor decides WHEN, the LLM decides WHAT and HOW" — a monitor-driven event pipeline that gives deterministic timing to non-deterministic LLM reasoning. Validated through an 8-flight Monte Carlo campaign (448 LLM calls, A- average grade) across STARLINK, NASA_CREW, USSF_CLASSIFIED, COMMERCIAL_GEO, and UNIVERSITY_CUBESAT mission profiles.

Multi-Agent LLM G-FOLD Monte Carlo Combustion Instability Python

Classical GNC Reference Simulator + Book

FletchGNC + Beyond Proportional Navigation

End-to-end classical GNC stack and a 44-chapter self-authored technical book

Reference simulator: 9-state Extended Kalman Filter navigation at 1 kHz, Proportional Navigation guidance at 100 Hz, plant-inversion control at 1 kHz on RK4 1 ms physics; quaternion attitude with Rodrigues integration; tactical-grade MEMS IMU model (BMI088-class) with Gauss-Markov bias, scale factor, cross-axis misalignment, G-sensitivity. ORBITAL_GNC_REFERENCE.md documents the architecture as a blueprint for a missile-seeker / orbital-rocket successor.

Core principle: author the canonical reference, then live by it. The companion book takes a reader from the geometry of pursuit, through classical and optimal PN, constrained / predictive / sliding-mode / impact-angle guidance, modern nonlinear control, low-Re aerodynamics and terminal ballistics, into symplectic, information-geometric, octonionic, Lie-group, and Koopman-spectral guidance — and the Monte Carlo validation that says when you can trust any of it.

9-state EKF Proportional Navigation Plant Inversion RK4 1 kHz Python

LLM-Driven Integration Air-Gap Capable

Universal Interface

LLM-driven middleware for federal legacy-system friction, with air-gapped classified routing

Discovers heterogeneous system I/O (REST / OpenAPI, CSV, JSON, PDF AcroForm, HTML form, SQL via any SQLAlchemy dialect, screen-scrape stub); LLM-generates a single-file Python connector; auto-synthesizes pytest-style cases (schema-derived, edge, PII-aware adversarial); iteratively refines until tests converge; logs sessions / token cost / latency / connector versions in SQLite telemetry. AST-policy + subprocess sandbox isolation on every generated connector.

Core principle: instead of one fragile bespoke connector per system pair, build a harness that builds, tests, and operates them on demand — and route by data sensitivity. Grok 4.3 for unclassified work; Qwen3 via Ollama, local and air-gapped, for restricted / classified — no manual switching. Cited problem: 80% of federal IT spend on legacy maintenance; 53% of agencies still on manual data transfer.

Grok 4.3 Qwen3 (Ollama) AST Sandbox Air-Gapped SQLite Telemetry

Multi-Agent Training CJIS-Prep · Officer Pilot

CopApp.ai

Multi-agent LLM training platform for police de-escalation

Three-agent pipeline (Dialog → World → Psyche) drives scenario simulation, automated grading on an ICAT-aligned rubric (6 categories / 10 points), 10-dimension session-insight scoring, skill-tree progression, and Bronze → Diamond certification tiers with freshness decay. Five specialized training UIs (Reps, Hidden Secret, Action-Detraction, Trainer, Tactician, Dojo). Department management, ZIP-based community discovery (Mapbox), Stripe Solo $9 / Department $29-per-seat billing.

Core principle: bring high-reliability-organization training rigor to a decision-making domain that doesn't have it yet, then make the CJIS-prep posture a non-negotiable from the rules layer up. 30-minute inactivity timeout, auth-event audit logging, CJI upload disclaimers, hardened Firestore rules with privilege-escalation checks. 60 commits / 14,000+ lines TypeScript / 54 cloud functions / 22 routes.

3-Agent LLM Pipeline ICAT Rubric Firebase Functions v2 OIDC + PKCE Stripe CJIS Posture

Voice-First / Mobile Sworn-Handler Beta

PawTrek

"The K9 Handler's Unfair Advantage" — AI-first platform for police K9 units

Voice-first capture (Deepgram + Grok) generates court-admissible narratives aligned with Florida v. Harris admissibility expectations; case-law verification; offline-first SQLite mirror with Firestore outbox and per-collection watermarks. Turborepo: React Native 0.83 / Expo SDK 55 (New Architecture, React 19) mobile, Next.js 16 web, Firebase Functions v2 backend, shared TypeScript domain spine.

Core principle: turn paper records and form-filling apps into a substrate-first system where the handler talks and the system organizes — health and wellness, skills and gaps, deployments, trainings, case-law context, vet portals, professional email CCs. CMEK enabled on Firestore from project creation; App Check with real attestation providers (App Attest / Play Integrity / reCAPTCHA Enterprise); CJIS posture documented. V5.0 design with active 25-finding hardening sweep, beta with a sworn handler in Monroe County, AL.

Deepgram Voice Grok 4.3 Expo SDK 55 Next.js 16 CMEK + App Check Offline-First

Aerospace & Research Work

NASA Artemis

SLS Block 1B Universal Stage Adapter

Systems engineer at Beyond Gravity (2022-2025) for the composite structure connecting Orion to the rocket for Artemis 4+. Owned requirements integration, verification, human systems safety, and compliance with NASA human-rating standards. Managed risks, navigated DoD/NASA specifications (DOORS, MIL standards), coordinated hazard closure through test artifacts. Heavy use of Python for engineering analysis.

Human-Rated Spaceflight Systems Engineering Safety & Risk CFRP

International

Payload Separation System (Sweden)

Spent 12 weeks in Linköping, Sweden working on NASA Exploration Upper Stage subsystem. Resolved NASA-standard compliance deadlock between Beyond Gravity and Leidos-Dynetics prime. Identified production lifecycle improvements (material selection, fastener methodology). Reviewed incident data, determined risk levels, coordinated mitigations, and closed hazard records.

NASA Standards Prime Integration Process Improvement

Research

Complexity in High-Reliability Organizations

Published research at AIAA SciTech 2022 ("Managing Complexity through Quantifying Communication in HROs") and INCOSE Detroit 2021 ("Managing Complexity through Collaborative Intelligence"). Developed frameworks to quantify communication patterns and collaborative intelligence in aerospace teams. Goal: find deep patterns in language and behavior that enable positive control of emergent properties in complex systems.

HRO Theory Communication Analysis Systems Thinking

Cal Poly

Spacecraft Design & Hyperloop

Senior Capstone (2020-2021): Led C&DH and ground systems for interstellar object intercept mission. Designed ground comms strategy for 7 AU range, 100 kbps downlink. Hyperloop (2018-2019): Top 22 worldwide at SpaceX competition. Authored test manuals, ran braking experiments (MATLAB), heat transfer analysis for 80mph→0 in under 8 seconds.

Mission Design Test Engineering Team Leadership

Curious about other things I've built?

View side projects & experiments

Publications

Research on complexity management and high-reliability organizations in aerospace contexts.

View Google Scholar Profile

2022 Conference Paper

Managing Complexity through Quantifying Communication in High Reliability Organizations

AIAA SciTech Forum, January 2022 (Paper #2022-0994)

Presents a quantitative framework for analyzing communication patterns in high-reliability organizations. Explores how measuring communication structure and flow enables teams to identify failure modes before they manifest, improving operational safety in aerospace and other complex technical environments. Anticipates today's AI-augmented engineering loops.

Read Paper

2021 Conference Paper

Managing Complexity through Collaborative Intelligence

INCOSE International Symposium, Detroit, November 2021 (Paper #129)

Examines how collaborative intelligence frameworks enable organizations to navigate complex system challenges through distributed cognition and coordinated action. Proposes novel approaches to enhancing operational efficiency in aerospace by understanding and leveraging emergent team properties.

Read Paper

2025 Technical Book · 44 Chapters

Beyond Proportional Navigation

Self-published technical reference accompanying the FletchGNC project

End-to-end reference on guidance, navigation, and control for guided projectiles. Geometry of pursuit → classical & optimal Proportional Navigation → constrained / predictive / sliding-mode / impact-angle-time / energy-optimal guidance → inertial navigation, Kalman and nonlinear filtering → NDI / adaptive control / actuator allocation → low-Re aero, Magnus, terminal ballistics → symplectic, information-geometric, octonionic, Lie-group, and Koopman-spectral guidance → Monte Carlo / testing / validation. Pairs with a working reference simulator.

Project Write-up Repository

2025 Mathematical Research · In Review

Wheel-Laurent Algebra: A Rigorous Formalization

Peer-review-grade synthesis with Python computational verification suite

Rigorously synthesizes Carlström's wheel-of-fractions with formal Laurent series. Identifies a small but genuine contribution — a total symbolic algebra distinguishing four types of zero with provenance carried through indeterminate forms — and explicitly refutes prior overclaims on Yang-Mills resolution. Python verification suite covers Cayley-Dickson, sedenions, Puiseux, and Borel summation.

2026 Working Draft · 13-Paper Collection · 241 pages · Not Yet Circulated for Peer Review

A Non-Associative Algebraic Framework for Yang–Mills Existence and Mass Gap

Self-compiled collection · Independent Researcher, Huntsville, AL · February 2026

Unverified rev 1 of a working collection of 13 papers exploring whether the Yang–Mills mass gap can be approached through a non-associative algebraic framework grounded in imaginary octonions and Malcev algebras. Opens with a Jacobi-Alternativity No-Go Theorem showing that embedding a Lie algebra into an alternative algebra cannot introduce genuine non-associativity — motivating the need for Im(O) as the external source. Builds through a contextual Poincaré–Birkhoff–Witt theorem, a positive-definite decompactified Killing form, contextual octonionic algebras (G₂-covariant non-associative deformation of su(3)), Moufang–Kolmogorov spectral theory, tree-filtered Sobolev estimates, and coherence conservation as a superselection principle from non-associativity. Constructs a lattice gauge-scalar theory with octonionic field values (reflection positivity, continuum limit), argues a Feshbach–Schur mass gap for octonionic gauge-scalar theories, Φ-marginalization preserving Wilson-loop correlators, pure G₂ Yang–Mills in seven dimensions, and culminates in a resolution paper. Consistent with — and complementary to — the Wheel-Laurent work, which specifically refutes prior overclaims. Status: unverified working draft, hedged accordingly. The math wants the same rigor I held others to.

Read Working Draft (PDF, 241 pp)

2024 Working Paper

The Patterns of Police Communication As It Relates to Simulated Behavioural Interaction with Procedural AI-Generated Non-Playable Characters In A Text-Based Environment

Working paper underpinning CopApp.ai's three-agent training architecture

Maps the structural patterns of officer-civilian communication in tense encounters onto a procedural AI-NPC scenario substrate. Connects PERF / ICAT-aligned outcome rubrics to agent-side state machines and dialog generation prompts; underpins the Dialog → World → Psyche pipeline now in the CopApp training platform.

CopApp Write-up

Areas of Expertise

Aerospace Systems Engineering

Requirements derivation / allocation / documentation; V&V plan development; IBM DOORS (heavy); CAMEO / MBSE / SysML; functional analysis, system trade studies; risk & opportunity assessment; IPT coordination; compliance under NASA, MIL-STD, Army, Space Force, and human-rating standards; change / waiver / exemption disposition.

Propulsion, GNC & Simulation

Solid-motor adapter integration, payload separation systems, human-rating, CFRP production; classical GNC stacks (Proportional / Augmented PN, ZEM-ZEV optimal guidance, NDI control, sliding-mode, plant-inversion); G-FOLD lossless-convex landing guidance; combustion instability (CHUG / L1 / T1); 9-state EKF; Monte Carlo trajectory & UQ; powered explicit guidance; thrust vectoring; RCS allocation.

Applied AI / Agentic Systems

Multi-agent LLM orchestration (Claude, Grok, GPT) with monitor-driven event pipelines; air-gapped local-LLM routing for classified data (Qwen3 via Ollama) with cloud-LLM fallback; agentic codegen (Claude Code, Cursor); LLM-driven schema discovery and connector generation; AST-policy sandboxing; real-time voice agents (Deepgram + xAI); RAG over million-token corpora; MCP / tool-use integration.

Production Security & Compliance

OIDC + PKCE, JWKS validation, secure cookie sessions, refresh-token rotation with reuse detection; token-bucket rate limiting; idempotency keys; HMAC + timestamp/nonce replay protection; CMEK on Firestore from project creation; App Check with real attestation providers (App Attest / Play Integrity / reCAPTCHA Enterprise); CJIS-prep audit logging; hardened security rules with privilege-escalation checks.

Fullstack / Cloud Production

TypeScript / React / Next.js / React Native (Expo); Node.js / Firebase Functions v2; Firestore (multi-tenant, hardened rules), Storage, Hosting; Stripe Checkout + Webhooks; Turborepo monorepos with shared TS domain spine; offline-first SQLite mirror with Firestore outbox; Pub/Sub queues with circuit breakers and jitter retries.

Mathematical & Research Depth

AIAA + INCOSE published in complexity / HRO. Author of Beyond Proportional Navigation (44 chapters) and Wheel-Laurent Algebra: A Rigorous Formalization (in review). Comfortable with symplectic, information-geometric, octonionic / sedenionic, Lie-group, and Koopman-spectral methods — and equally comfortable refuting overclaimed results when the math doesn't hold.

Resume

Comprehensive overview of experience, education, and technical capabilities.

Download PDF

Resume — Alexander Derkatsch

Get In Touch

Exploring opportunities at the intersection of domain expertise and AI orchestration. Open to conversations about:

AI-augmented aerospace roles: Orchestrating AI for simulation, design optimization, requirements analysis, or safety-critical systems in aerospace/defense

Hybrid AI product roles: "AI orchestrator," "AI systems architect," or emerging titles combining domain expertise + AI workflow design + product building

CopApp partnership/investment: AI-powered training platform for high-stakes decision-making in power-imbalanced contexts (law enforcement, emergency response)

Complex systems + AI: Roles bridging HRO theory, safety-critical design, and AI/ML integration for organizations managing high-consequence operations

ai.derkatsch@gmail.com alexi@copapp.ai +1 (609) 815-0257 @AlexiDerkatsch

Experience

Founder / Applied AI Systems Engineer

Aerospace Systems Engineer

Research — Complexity in High-Reliability Organizations

Hyperloop Team — Braking Systems

Flight Coordinator

Education

B.S. Aerospace Engineering

Private Pilot License