operational workflows in one command center
Integrated 21 operational workflows into one AI-assisted command center, consolidating coordination, exception visibility and process monitoring so handoffs stop depending on manual follow-up.
Hello, I’m
APIs, webhooks and workflow automation that stay correct under duplicates, timeouts and retries.
HOW I BUILD
Reliable APIs, explicit state, and recovery paths. Not just the happy path.
See systems and proof →CLIENT WORK
Integrated 21 operational workflows into one AI-assisted command center, consolidating coordination, exception visibility and process monitoring so handoffs stop depending on manual follow-up.
Built n8n workflows connecting LinkedIn, email and the client website for lead generation, outreach and follow-up, contributing to approximately 2x lead volume.
SELECTED WORK
Lead intake, documents, and controlled AI.
A working Next.js interface, Fastify API and PostgreSQL worker system. Configure workflows, approve execution, inspect saved attempts, and recover failed runs.
58/58 system tests · 10/10 browser cases in each native and Docker environment · 10/10 Docker smoke checks
Deterministic HTTP sandbox for AI triage; no live production LLM or customer deployment.
Next.js · React · TypeScript · Fastify · PostgreSQL · Prisma · Docker

AI actions need boundaries and safe recovery. This reference records intent before API calls, holds sensitive writes for approval, and reconciles uncertain outcomes.
261 tests · 90.01% coverage · preserved v2
Source & tests →Find answers in current documents. This reference processes files, tracks versions, returns source citations, and holds uncertain extraction for review.
134 tests · 91.64% coverage · 12/12 evals
Source & tests →Independent tested reference. Results are archived v2 evidence, not current commercial use or universal exactly-once delivery.
Default retrieval uses deterministic lexical embeddings and extractive answers. Twelve synthetic regression cases do not establish commercial semantic retrieval quality.
These are tested engineering references, not claims of live customer scale or uptime. Test figures describe the linked revisions. Agent evidence · Knowledge verification
ALSO BUILT
LOCAL LLM / AGENT ORCHESTRATION
Local-first AI coding agent with model adapters, durable dispatch, approval-gated code proposals and GitHub publication safeguards. Real CPU-based local inference verified; isolated code execution remains disabled pending security validation.
TYPESCRIPT / POSTGRESQL
Lead intake with signed webhooks, duplicate-event checks, bounded retries, and saved handoff outcomes.
N8N / JAVASCRIPT
Lead routing and follow-up controls: suppression, state checks, no-send paths, and recovery evidence.
LOCAL-FIRST AI AGENT / PYTHON / SQLITE
Experimental Windows AI workspace with owner-approved LLM request routing, durable memory and task state, scoped read-only tools and a crash-aware execution journal. CI on Windows and Ubuntu.
Also: BIMMCA Intelligence ↗: Supabase browser consumer.
CAPABILITIES
Python · FastAPI
PostgreSQL · Pydantic
n8n · REST APIs
Webhooks · TypeScript
MCP · Retrieval
Structured outputs · OCR
Idempotency · Approvals
Tests · CI · Recovery
CONTACT
Backend & Integration Engineer working in Python, FastAPI, TypeScript and PostgreSQL. Open to remote full-time and contract roles.