I turn ambiguous, high-stakes systems — AI, money, real-time — into boring, predictable infrastructure.
I build systems where AI carries real production load—and I have spent 15+ years shipping backends and platforms that stay predictable, observable, and operable when usage grows.
AgentStack.tech is a patent-pending, AI-first Universal Processing System: full backend, admin surfaces, and a unified integration plane—TypeScript/Python/React SDK, REST, and MCP with 550+ catalog actions spanning projects, auth, RBAC, billing, wallets & buffs, scheduler, webhooks, analytics, AgentSocial, SSE messenger, RAG & agent memory, CRM, commerce & marketplace, bots & integrations, an on-chain AgentNet economy, and server-side logic.
Role: Founder & Chief Architect—product architecture, protocol & SDK design, multi-tenant data model, AI instruction systems, and end-to-end delivery (dual-shell SPA, TanStack Query, SSE + WebSockets, PostgreSQL, Stripe, Solidity/EVM, observability). Through-line: performance discipline from AAA-style game engineering applied to AI-native infrastructure.
AgentStack is designed so humans and AI agents share the same contract: typed SDK, REST, and MCP—with discovery, documented workflows, and governance baked in—not bolted on.
Most “AI backends” glue an LLM to CRUD. AgentStack is inverted: a multi-tenant universal processing core with Field Access Policy, Protein Command System, Logic Engine, RAG, commerce, social, realtime, and an on-chain AgentNet economy—then a first-class MCP + REST catalog (550+ actions) so agents can operate the same platform developers do.
Connect time for new products is measured in minutes because migrations, rollback, RBAC, billing, and observability are part of the architecture—not weekend glue code.
Service orchestration with a universal DNA data model—same conceptual backbone across SaaS, games, social, and commerce verticals.
O(1)-style routing with aggressive caching philosophy—documented target <0.001s on the router hot path; built for horizontal scale.
Universal command bus aligned with automation, internal tooling, and MCP-shaped execution—one vocabulary for scripts and agents.
Role- and user-scoped field masks and triggers—secure exposure for REST, SDK, and LLM-mediated access.
Server-side event-driven rules and processors—business logic without shipping a new microservice for every branch.
agentstack.execute with 550+ actions, discovery, health, and bundled AI prompts (workflows + use-case packs).
Separate user vs developer/admin experiences (mobile-first), TanStack Query server state, 600+ component modules, i18n.
SSE for messenger & social streams; WebSockets for operational/build telemetry—clean separation of product vs ops channels.
How AI is constrained to behave well in production—not only which model you call.
Great AI systems are interfaces + data + policy. Models change weekly; contracts and governance stay.
ai_prompt bundles and use-case packs (game, SaaS, ecommerce, social).Quantified engineering outcomes where documented—plus shipped scope that demonstrates architectural ownership. Beyond features: architecture decisions are written down, and invariants are enforced by CI — not by memory.
These figures appear in AgentStack engineering documentation as consolidation and neural-architecture targets—they summarize intentional core unification, not incidental tuning.
| Dimension | Outcome | Notes |
|---|---|---|
| Core consolidation | ~87% fewer services | Documented reduction (15 → 2 primary core services) to cut coordination and ops drag. |
| Memory footprint | ~76% reduction | Documented unified-core target (822MB → 200MB class targets in architecture docs). |
| Neural Router | <0.001s · O(1) routing | Hot-path latency budget; paired with neural cache discipline. |
| Neural events | 200,000+ events/s | Throughput class for event-driven orchestration paths. |
| Neural cache | 95%+ hit-rate target | Designed into routing and read-heavy platform paths. |
| MCP surface | 550+ actions | Single discovery-oriented catalog for agents—same platform as humans. |
| Product UI kit | 600+ component modules · i18n | Shipping velocity for admin + dashboard surfaces (`agentstack-frontend/src/components/`). |
| Architecture Decision Records | 375 ADRs | One documented decision per major subsystem — `docs/adr/`, not tribal knowledge. |
| Architecture-invariant CI audits | 123 audit scripts | Automated boundary checks (shell parity, SEO surface, hosting golden paths) beyond unit tests. |
| End-to-end test coverage | 118 Playwright specs | Hosting, commerce, CRM, bots, integrations, security flows. |
Figures cited from internal/platform engineering documentation (architecture guides, neural router targets) and direct repository counts (ADRs, CI scripts, e2e specs), verified on the date noted in each row's source comment. Treat as engineering claims tied to design intent—always happy to discuss methodology on a call.
Codebase scale: 3,600+ Python modules (agentstack-core/) and 4,000+ TypeScript/TSX modules (agentstack-frontend/src/) — file counts, not line counts, as the more defensible metric.
Live, checkable proof — not just claims:
Continuous ownership, not a one-time build: shipped major subsystems on top of the original core.
domain.subsystem.role.genN) that lets AI coding agents locate the right module tree in a large codebase without full-repo search — later exported as a portable starter kit for other projects.Faced with LLM-generated code breaking on templating-syntax collisions across products, instead of patching each occurrence, redesigned the underlying command/template contract so the conflict class became structurally impossible — a pattern later reused across the AI Builder pipeline's validation gates.
Backend (FastAPI, Pydantic, SQLAlchemy, 8DNA / Neural Router patterns, in-process work queue & cache — no Redis/Kafka dependency), frontend (React, Vite, Tailwind, dual-shell SPA, TanStack Query), data (PostgreSQL), LLM application architecture (MCP catalog, custom RAG engine, agent memory, agentic workflows), commerce (Stripe, wallets, buffs), real-time (SSE messenger, WebSocket ops), Web3 (Solidity, multi-chain testnet orchestration). Observability: Loguru, OpenTelemetry. Unified SDK (TypeScript @agentstack/sdk, React hooks). Cloud-native, multi-tenant, event-driven.
FastAPI, Data processing, AI/ML integration, Automation
React, Node.js, Real-time systems
Game development, Backend services, ASP.NET Core
Enterprise architecture, Microservices, System design
PostgreSQL, Complex queries, Database optimization
REST APIs, async/await, MCP server, AgentStack core
ASGI server, WebSocket support, production deployment
WebSocket: build status, dashboards. SSE: messenger & real-time social (AgentStack)
Data validation, settings, API schemas (v2)
Event-driven architecture, Real-time communication
Enterprise applications, Microservices
Rapid development, Data processing (pre-AgentStack projects)
Hooks, concurrent features, dashboard & builder UI
Strict mode, SDK types, frontend and tooling
Build tooling, HMR, React plugin, base path
SPA routing, nested routes, loaders
Server state, caching, invalidation (AgentStack)
Utility-first, design tokens, responsive UI
Node-based UIs, logic builder, diagrams
Internationalization, locale switching
@agentstack/sdk, React package, protein-style API consumption
Frontend unit/integration tests, Vite-native runner (AgentStack)
ORM, migrations, async engine, AgentStack core
Async PostgreSQL driver, high performance
Async HTTP client, outbound API calls
Schema design, Complex queries, Performance tuning, JSONB
In-process queue/cache on unified DNA table (AgentStack) — chosen over Redis/Kafka to cut ops surface; Redis used in earlier enterprise/game-backend roles for caching & pub/sub
Document storage, flexible schemas (pre-AgentStack projects; AgentStack itself is PostgreSQL-native)
Real-time databases, authentication, cloud functions (pre-AgentStack / indie projects)
Full-text search, log aggregation (pre-AgentStack projects)
Authentication, token validation, password hashing
Payment gateway, subscriptions, AgentStack billing
API keys, sensitive data, TLS
Structured logging, rotation, AgentStack core
Traces, instrumentation (FastAPI, HTTP)
Backend and frontend testing, async tests
Code quality, type-check, build pipeline
Orchestration of LLMs and agents, multi-agent patterns, MCP at scale (550+ catalog actions), discovery, workflow docs, use-case prompts (game, SaaS, ecommerce, social). Custom-built RAG engine (self-authored vector search & compression, no third-party vector DB), collections, semantic search, agent long-term memory. AI instruction design: system prompt bundles, structured workflows (understand → plan → build → publish), prevention-over-cure. Providers: OpenAI, Anthropic, local LLMs (Ollama, vLLM). Orchestration: self-built multi-stage pipeline (AgentStack AI Builder); LangChain/LangGraph evaluated in earlier prototypes, not in AgentStack production.
API integration, function calling, structured outputs
Local LLM deployment, model management, self-hosted
Self-built multi-stage pipeline (understand → plan → build → publish) with gate/repair-loop validation — not LangChain/LangGraph; those were evaluated in earlier, pre-AgentStack prototypes
Self-built vector engine — Fast Walsh-Hadamard rotation + outlier-aware quantization (HGT algorithm from Google Research/DeepMind, ICLR 2026), ~8× compression, no third-party vector DB dependency
agentstack.execute, 550+ actions, discovery, AI prompt bundle
Knowledge-augmented generation, retrieval, evals
System prompts, workflows, MCP ai_prompt / use-case bundles, AI Builder
Designed and shipped an on-chain economy layer (AgentNet) with an append-only Postgres ledger as financial source of truth, EVM smart contracts as public anchors, and multi-chain test-net orchestration — not a token bolted onto a database.
Status: testnet / pre-mainnet evidence stage (Base Sepolia, BSC testnet, Arbitrum Sepolia). Engineering framing, not a financial offering: this section describes vault accounting mechanics and a reference implementation, not investment returns.
ERC-4626 yield vault (agUSD over USDT), ERC-8004 execution-proof registries, EIP-712 signed envelopes, Timelock governance
Base Sepolia, BSC testnet, Arbitrum Sepolia via a single ChainProfileRegistry; faucet + scenario orchestrator
Merkle checkpoints, hash-chained ledger batches, public verifiability without a shadow source of truth
Rate-limited, finality-aware mint/burn bridge policy; AGNT (utility) vs agUSD (vault shares) vs USDT payments — three separated rails by design
Live evidence lab: agentstack.tech/grants/demo — public multi-chain proof timeline.
Verified testnet contract (AGNT, Base Sepolia) — cross-checked against docs/grants/evidence/grant_audit_pack_latest.json, 2026-07-07:
sepolia.basescan.org/address/0x38aD…9a943 — source-verified, compiler v0.8.24. Testnet contracts may be redeployed; re-verify before relying on this address.
Containerization, Image optimization, Multi-container setup
Orchestration, Scaling, Deployment
EC2, Lambda, RDS, S3, CloudFront, API Gateway
Version control, CI/CD, Complex workflows
Aligned with shipped platform work — multi-tenant core, MCP catalog, social & messenger (SSE), RAG, billing, unified SDK, Web3/AgentNet, observability.
Core competencies
| Skill | Level | Experience | Context |
|---|---|---|---|
| System Architecture Design | Expert | 10+ years | Game servers, microservices, distributed systems |
| Microservices / multi-tenant SaaS | Advanced | 8+ years | Per-tenant clusters, migrations, AgentStack production |
| API Design (REST/GraphQL) | Expert | 10+ years | RESTful principles, GraphQL implementation |
| Database Schema Design | Advanced | 10+ years | PostgreSQL, MongoDB, normalization, optimization |
| Real-time Systems Architecture | Expert | 8+ years | WebSocket, multiplayer backends, live data, AgentStack build status |
| Pydantic / API Schemas | Advanced | 3+ years | Validation, settings, OpenAPI, AgentStack core |
| SQLAlchemy (async) | Advanced | 3+ years | ORM, migrations, PostgreSQL, AgentStack |
| Field Access Policy (FAP) | Advanced | 2+ years | Role/user field masks, triggers, secure data exposure |
| Protein command bus / DNA ops | Advanced | 2+ years | Universal automation surface, integrations |
| Technology | Level | Experience | Context |
|---|---|---|---|
| React (hooks, concurrent) | Advanced | 5+ years | Dashboard, builder UI, AgentStack frontend |
| TypeScript | Advanced | 6+ years | Strict mode, SDK types, frontend and tooling |
| Vite | Advanced | 2+ years | Build, HMR, React plugin, AgentStack |
| TanStack Query | Advanced | 2+ years | Server state, caching, AgentStack |
| Tailwind CSS | Advanced | 3+ years | Utility-first, responsive UI |
| React Flow | Advanced | 2+ years | Node-based UIs, logic builder |
| Dual-shell SPA / audience routing | Advanced | 1+ years | User vs dev surfaces, mobile-first admin (AgentStack) |
| SSE & real-time UX | Advanced | 2+ years | Messenger streams, social presence patterns |
| Language | Level | Years | Current Use |
|---|---|---|---|
| Python | Advanced | 5+ | FastAPI, Pydantic, AgentStack core, AI/ML |
| TypeScript/JavaScript | Advanced | 6+ | React, Vite, TanStack Query, AgentStack frontend |
| C# | Expert | 10+ | Legacy game dev, ASP.NET |
| Java | Advanced | 5+ | Enterprise systems, Architecture |
| SQL | Expert | 10+ | Complex queries, optimization |
| Technology | Level | Experience | Specialization |
|---|---|---|---|
| Ollama | Advanced | 2+ years | Local LLM deployment, model management |
| LangChain / LangGraph | Advanced | 2+ years | LLM chains, prompt engineering, memory — evaluated in earlier prototypes, not in AgentStack production (AgentStack uses a custom orchestrator, see below) |
| Custom RAG Engine (TurboQuant) | Advanced | 2+ years | Self-built vector search — Fast Walsh-Hadamard rotation + outlier-aware quantization, ~8× compression; no third-party vector DB |
| MCP Protocol | Expert | 2+ years | 550+ actions, discovery, workflow & use-case prompts |
| RAG Systems | Advanced | 2+ years | Collections, semantic search, agent memory (production) |
| Custom AI Builder orchestrator | Advanced | 2+ years | Multi-stage pipeline (understand → plan → build → publish), gate/repair-loop, function calling |
| Technology | Level | Experience | Context |
|---|---|---|---|
| Solidity / EVM contracts | Advanced | 1+ years | ERC-4626 vault, ERC-8004 proof registries, EIP-712, Timelock governance (AgentNet) |
| Multi-chain orchestration | Advanced | 1+ years | Base Sepolia, BSC testnet, Arbitrum Sepolia via ChainProfileRegistry |
| On-chain anchoring & ledger design | Advanced | 1+ years | Merkle checkpoints, hash-chained batches, Postgres ledger as source of truth |
| Bridge & token economics design | Advanced | 1+ years | Rate-limited mint/burn policy, AGNT/agUSD/USDT rail separation |
| Area | Level | Experience | Context |
|---|---|---|---|
| JWT / bcrypt / Auth | Expert | 8+ years | Token validation, password hashing, AgentStack |
| Stripe / wallets / buffs | Advanced | 2+ years | Subscriptions, trials, multi-wallet economy |
| Loguru / OpenTelemetry | Advanced | 2+ years | Structured logging, traces, AgentStack |
| pytest / Jest | Expert | 8+ years | Backend and frontend testing, async, AgentStack |
15+ years in backend and distributed systems; game development, full-stack, AI/LLM, and Web3. AgentStack — Universal Processing System: MCP + REST (550+ actions), social & messenger, RAG, commerce, AgentNet economy, unified SDK, dual-shell product UI.
Tech: Python, FastAPI, PostgreSQL, Docker, Compose, queues, Stripe and payment rails, Solidity/EVM. React, TypeScript, Vite, Vitest. MCP, RAG, WebSockets (ops), SSE (chat). Observability: Loguru, OpenTelemetry. IDE/agent plugins: Cursor, Claude, GPT, VS Code.
Dreamfrost Studio
Led development of Skybringer: Idle RPG—adaptive UI, core mechanics, SDK expansion—shipping on mobile and desktop-class targets.
World4Play
Toadman Interactive (Stockholm)
ArtPlant (Oslo)
Self-taught, portfolio-driven engineer — no formal CS degree. 15+ years of shipped production systems (games → distributed backends → AI infrastructure) as the primary credential; architecture decisions are documented in 375 ADRs rather than claimed by title.
Consulting and collaboration: AI systems architecture, MCP integration (large action catalogs), RAG & agent memory, multi-tenant BaaS, billing, real-time social, Web3/AgentNet — and AgentStack ecosystem.
AI Systems Architect · Full-Stack · Founder, AgentStack.tech
Availability
Remote-first · GMT-3 (Brazil, Santos) · Open to discussing new engagements
Location
Santos, São Paulo, BrazilPhone
+55 (13) 99800-4433Telegram
@LanceW4P