AI Infrastructure
Deploying custom Model Context Protocol (MCP) servers to seamlessly bridge enterprise APIs with intelligent workflows.
AI Infrastructure & B2B Solutions Engineer
David Sequera
I engineer high-performance web infrastructure and custom AI integrations. With over six years of experience across frontend architecture and systems development, I do not just write code. I design resilient systems that eliminate operational bottlenecks.
Available for systems workBased in Caracas · Remote
Selected Work
Four systems in production. Architecture choices and trade-offs annotated.
Core Capabilities
Frontend Architecture
Architecting efficient, scalable, and high-trust web platforms using the modern JavaScript/TypeScript ecosystem.
Interactive Engineering
Leveraging C# and Unity for logic-heavy, high-performance applications.
Case Studies
Three systems shipped against three operational bottlenecks. Problem, architecture, and direct business impact, annotated.
[01] AI Infrastructure · Real EstateLIVE
MLS_MCP
MLS MCP: The Live MLS, Callable by an AI Agent
A remote Model Context Protocol server that exposes the MiamiRE MLS as eight read-only tools an AI agent can call directly.
Uptime
99.9%
p50 latency
~280ms
Tools exposed
9
Stack
- Python
- MCP SDK
- FastAPI
- Pydantic
- MLS Vendor API
- Docker
01
The Bottleneck
An agent's day fractures between two systems that don't speak. The CRM holds buyer intent (tags, budgets, follow-up cadence) but no inventory. The MLS holds every active listing, comp, and open house, locked behind a feed no agent, and no AI, can query in the moment. The bridge between them is manual: toggle portals, copy MLS numbers, rebuild the same query, send the link hours after the lead has already cooled.
02
The ROI
A Follow Up Boss lead resolves to ranked live listings, each carrying a rationale and a send-ready Smart Move Miami link, in a single tool call. CMAs collapse from an afternoon of manual comps pulls to seconds. Read-only by construction and unattended-safe, so CRM automations call it without supervision. The MLS, finally callable.
The Architecture
A remote MCP server that exposes the live MiamiRE MLS as eight read-only tools: multi-filter search, lead-to-listing matching, comparables, fix-and-flip heuristics, open houses, address-to-link resolution. Sits between the agent runtime (Claude or a CRM-bound workflow) and the RESO Web API behind a hardened bridge adapter: TTL cache, per-agent auth, structured retryable error codes, field names verified against the live feed metadata. Every tool returns its result or a typed error (never throws), so it is safe to fire from automated triggers.
[02] AI Automation · Inside SalesLIVE
SMART_MOVE_MIAMI
Smart Move Miami: Your AI Lead Manager
An always-on inside sales agent wired into Follow Up Boss and Gmail: qualifying, drafting, and guarding the pipeline end-to-end.
Triage time
−92%
- Smart Move Miami stack
Stack
- Claude API
- MCP
- Follow Up Boss
- Python
- Postgres
- Webhook ingress
01
The Bottleneck
The first ten minutes decide the deal. A fresh inbound is interested right now; wait too long and the lead goes cold or calls whoever answered first. In an active agent's day, new leads sit unworked through showings, follow-ups vanish between calls and closings, and good leads quietly stale at the bottom of the list. Every missed minute is missed pipeline.
02
The ROI
Ten-minute response time on every inbound, 24/7. Zero leads forgotten. 100% of interactions logged, stage-correct, and followed up with a date-bound task. A weekly coaching pass surfaces stuck leads, neglected hot prospects, and the single highest-value action of the week. A full inside-sales operation (speed, follow-through, coaching) without adding headcount.
The Architecture
An always-on inside sales agent wired into the existing stack: Follow Up Boss as canonical pipeline state, Gmail as the real conversation history. Drives a nine-stage state machine (Lead → Attempted → Follow Up → Options → Interested → Showing → Hot Prospect → Pending → Closed) with a parallel nurture track for unresponsive contacts. Reads the entire lead history before drafting, matches tone in English or Spanish, enforces the 10-minute speed-to-lead reminder, and writes notes plus dated follow-up tasks back to the CRM after every interaction. Hard guardrails: never contacts Not-Interested or Trash, never fabricates data, never sends without human approval.
[03] Real-Time Systems · Mobile + WebSHIPPED
WICK
Wick: Candle Timer for Traders
A UTC-precise, cross-platform candle-close timer that fires on the second the market prints: iOS, Android, Web/PWA from a single codebase.
Notify drift
<50ms
Timeframes
8
- iOS · Android
Stack
- React Native
- Expo
- TypeScript
- Native Notifications
- UTC Math
01
The Bottleneck
Setups confirm on the candle close, not a second before. The cost of missing it is structural: late entries, worse fills, hours of screen fatigue spent watching a clock instead of trading the plan, and the second-guessing that follows. Existing alerting apps round to “about now” and drift over time, which on a one-minute scalp eliminates the entire signal.
02
The ROI
Exact timing without screen fatigue. Set once, walk away. Every confirmation arrives on the second the market prints the close. One build covers three platforms and addresses both a 24/7 crypto market and a $7.5T/day FX market. 100 active alerts free; a Pro tier layers per-symbol pins, quiet hours, and notification-level snooze actions on top.
The Architecture
A cross-platform candle-close timer (iOS, Android, Web/PWA) shipped from a single codebase. UTC-aligned interval math snaps every alert to the exact candle boundary the whole market sees: 1m, 5m, 15m, 30m, 1h, 4h. The scheduler re-arms the instant a notification fires, so alignment never drifts across sleep, wake, or reload. Native OS notifications fire in the background and on the lock screen, on time. Crypto pricing from Binance, Forex from TwelveData. Stack reminders per candle: at the close, or N seconds to minutes ahead.
My objective is straightforward: replace complex technical friction with clean, automated, and highly effective solutions.
Get in touch→