Download PDF →

Summary

NARVI is an applied-AI company — agent orchestration, AI safety engineering, and agentic security — founded and directed by Akeim Joseph, who designs, directs, and verifies every system rather than prompting them into existence. A.R.C.H.O.N. is a multi-agent orchestration system built around a biomimetic safety architecture — bounded, non-agentic “Nano-AI” processes coordinated by a supervising agent, hard permission gates on every state-changing action, and a deterministic evaluation harness that grades agent-loop behavior against fixed test cases plus an LLM-judged quality score. Anubis is NARVI’s agentic-security product — a stateless compliance gate judging AI-agent actions against hard, auditable invariants before they execute. SFT Command was NARVI’s earlier financial intelligence platform — 200+ backend services, live fintech integrations (Plaid, Stripe, Yahoo Finance); after 8 months of self-funded outreach to fiduciary buyers produced zero pilots, that product was stopped rather than built past a market that wasn’t there — the entity itself continued, renamed to NARVI LLC. Nine years of direct customer-facing work — sales, service, high-throughput quality control — shapes how Akeim scopes and delivers technical systems: discovery first, architecture second, implementation directed and verified, not guessed at. NARVI is now looking for a cofounder with real enterprise sales experience — the systems are real and working; closing deals is the one piece Akeim has proven, on his own dime, that he can’t do alone.

Experience

Founder & Director — NARVI LLC August 2025 – Present · Augusta, GA

NARVI is the parent company behind A.R.C.H.O.N., Anubis, and PRISM — originally filed as SFT Command LLC, renamed once its first product (below) was stopped. Founded and directed solo; every architecture decision, every line of production code, and every trained model checkpoint across the portfolio is built and verified by Akeim directly.

Creator & AI Product Engineer — A.R.C.H.O.N. (NARVI, Independent R&D) August 2025 – Present · Augusta, GA

A multi-agent orchestration system: a supervising agent coordinates bounded, single-function “Nano-AI” processes under hard safety limits.

  • Nano-AI Safety Architecture — bounded, single-function, non-agentic narrow processes that each perform one deterministic task and connect only to explicitly wired data sources.
  • Evaluation Harness — deterministic assertions plus an LLM-judged quality score, grading agent-loop directives against fixed test cases.
  • Observability & Permission Gating — a live diagnostic panel exposing real-time node state and inter-agent messages, paired with a hard permission gate requiring explicit human sign-off before any autonomous code-writing or state-changing action.
  • Memory & Knowledge Architecture — a multi-layer memory system: working buffer, persistent memory (confidence-weighted RAG), and immutable historical memory.
  • Security Architecture — a 5-layer defense model with threat tracking, multi-code access verification, clearance levels, and authenticated encryption for node-to-node communication.

(Private repository — code available for review upon request)

AI Product Engineer — Anubis (NARVI, Private R&D) October 2026 – Present · Augusta, GA

A domain-scoped AI classifier and deterministic compliance gate for agentic systems — separate from A.R.C.H.O.N. by design, built to wrap any agentic system, not just NARVI’s own.

  • Invariant gate — judges a proposed AI-agent action by its real resulting state (resource sensitivity, origin, category before and after), not its stated intent; stress-tested directly against a concrete multi-step exfiltration scenario before the schema was finalized.
  • Classifier core — a purpose-built transformer (not a wrapped LLM) scoring CSA’s Agentic AI Autonomy Levels (L0–L5) plus resource sensitivity, origin, and category.
  • Real training discipline — every label is human-reviewed before it counts as training data; synthetic data is used only to prove the pipeline/architecture at scale, and is always labeled and logged as such, never conflated with real reviewed capability.

(Private repository — code available for review upon request)

Founder — SFT Command LLC (financial platform for independent RIAs) Building since August 2025 · LLC registered January 2026 · Augusta, GA (renamed to NARVI LLC)

  • Built a 200+ backend-function platform with live Plaid, Stripe, and Yahoo Finance integrations — 50+ database schemas, multi-tenant isolation, RBAC, compliance audit trails.
  • Architected five domain-scoped AI agents coordinated into a single compiled answer, plus predictive analytics (5-year net worth projection, cash flow forecasting, Monte Carlo simulation).
  • Eight months of direct outreach to fiduciary buyers produced zero pilots — a distribution and trust-network problem, not a technical one. Made the call to stop pursuing that product rather than keep funding a sales motion I’m not suited for. The LLC itself wasn’t dissolved — it was renamed and became NARVI LLC, the parent entity behind A.R.C.H.O.N., Anubis, and this site; code from SFT Command is retained as capability evidence.

Customer-Facing & Operations Experience

Nine years of direct customer discovery, technical product matching, and high-throughput quality control — the same discovery-first discipline now applied to scoping AI systems.

  • Customer Service — AutoZone · March 2024 – July 2024 · Louisville, KY
  • Busser / Server Assistant — Vincenzo’s · 2023 · Louisville, KY
  • Sales Associate — Middle of Town Cycling · August 2023 · Louisville, KY
  • Delivery Driver — Jimmy John’s · January 2023 – May 2023 · Louisville, KY
  • Line Associate — Smithfield Packaging · 2020 · NC
  • Busser / Dishwasher — Cracker Barrel · 2019

Additional prior roles available on request.

Education

Southern New Hampshire University (SNHU) — Online Bachelor of Science, Computer Science — In progress (transferred from Accounting)

Abilene High School — Abilene, TX High School Diploma with Criminal Justice Endorsement

Technical Skills

AI & Agent Systems — AI coding agent orchestration (Claude Code, Cursor, Base44), multi-model LLM routing, autonomous agent loops with bounded iteration limits, non-agentic narrow-task orchestration, deterministic evaluation harnesses, RAG pipelines, multi-agent task delegation with human-approval gating, voice AI, applied ML model training (PyTorch transformer classifiers, multi-head single- and multi-label architectures), reinforcement learning (PPO via Stable-Baselines3, custom Gym environments), real capacity-ceiling diagnostics separating learning gaps from architectural limits.

Languages — JavaScript (ES6+) and TypeScript, Python, SQL (Postgres/Supabase).

Backend Engineering — REST API design, microservices architecture (200+ deployed functions), webhook handling, event-driven systems, rate limiting, circuit breakers, audit logging, multi-tenant data isolation, RBAC, compliance ledger with hash-chain sealing, real crash-recovery and error-handling design (diagnosing a production connection-pool failure from its real stack trace and closing the actual gap, not just restarting the process).

Financial Technology — Plaid, Stripe, real-time market data, trading signal generation, Monte Carlo simulations, net worth projection, cash flow forecasting.

Frontend Engineering — React provider hierarchies, canvas-based visualization systems, Three.js 3D rendering, real-time data visualization, motion-budget animation systems.

Systems & Site Reliability — Jekyll/static-site architecture and content modeling (collections, layouts, templated data), GitHub Pages CI/CD, SCSS and CSS cascade debugging (diagnosing a real cross-version theme conflict down to the exact selector mismatch, not guessing at a fix), git branching and pull-request workflows across multiple repositories, local dev-server process management.