Bogor · Applied AI Engineer

Fudhoil

I use and manage AI inside real products — wire agents to tools, write skills that guide how models work, connect MCP to company systems, and put insights on live dashboards. I don’t invent foundation models; I make existing ones useful at work.

HashMicro Work

you@fudhoil ~ %

Practice

Applied AI — run it, wire it, keep it honest.

Not model research. I configure, integrate, and operate AI systems in production: agent platforms, skill libraries, MCP, product intelligence UIs, and delivery gates — on top of years shipping business software.

Use agents Configure LLM coworkers with tools, memory, channels, credits — plus topic-routed desks for coding / research / review
Manage skills + MCP HMX knowledge packs · skill hubs · PMT MCP evidence · OAuth tool servers
Product AI surfaces Dashboard Insight Brief · Explain with AI · contextual actions
Govern delivery PM → Investigator → Executor → QA · AI Score · runtime proof before ship
Operate models Serve / integrate APIs · put LLMs in real tools (e.g. spreadsheets)

Depth

How hard should the system think?

Same stack, different pressure — from everyday agent ops to full-auto delivery gates.

instant · everyday ops

Loop

Skills → MCP → Agent → Product

Skills load first — agents read cookbooks before touching code.

01 · Agents

I run in-product AI coworkers

Configure and operate LLM agents inside ERP — pick providers, prompts, tools, channels. The model is bought or plugged in; the system around it is what I own.

Provider plugs

Wire OpenAI-compatible + Anthropic APIs, model templates, identity prompts, secret scopes.

OpenAI · Anthropic · Templates

Tool-calling ops

Agent loops with memory, subagents, webhooks, and tool-loop guardrails — product agents in ERP plus a personal gateway for coding / research / review desks.

Loops · Guardrails · Subagents · Desks

Channels & credits

WhatsApp / Telegram sessions with topic routing (coding, research, review), participant scopes, usage events, company credit ledger.

Messaging · Topics · Sessions · Billing

02 · Skills & MCP

Steer models with rules and tools.

30+ HMX skill cookbooks

Shared knowledge pack for modules, XML views, Webx, security, reports, E2E, full-auto — agents load this before they touch ERP code.

SKILL.md · hmx-knowledge · Pattern discovery

MCP toolbelt

Stdio + remote MCP with OAuth where needed, secret substitution, reload. Web search/fetch for evidence, not vibes.

MCP · OAuth · Secrets · Web providers

PMT product context

Point agents at PRDs and tasks so answers come from roadmap truth, not vibes.

PMT MCP · PRD · Evidence

03 · Product intelligence

AI on the dashboard the user already sees.

AI Insight Brief

Shared host + store so dashboards register context once and get briefs without copy-paste plumbing.

Hashy · Insight Brief · Providers

Explain with AI

Per-widget explain actions from live snapshots — answers tied to visible data, not blank chat.

Contextual actions · Webx

Suggested actions

Validate/fallback AI suggestions so operators get next steps that actually make sense.

Action cards · Safety

04 · Delivery systems

Multi-agent pipelines with score gates.

Full-auto mode

Plan → implement → verify → commit → GitLab MR without asking every mundane step — safety rails kept.

Full-auto · MR · GitLab

AI Score reviewer

Independent gate scores evidence before push. Weak proof blocks ship.

Reviewer · Score ≥ 85 · Blockers

Role-separated agents

Product Manager · Investigator · Executor · QA — one agent doesn’t rubber-stamp itself. Done only with runtime proof against acceptance criteria.

Change workflow · Runtime QA · E2E

05 · Use & operate

Plug models in. Serve what we need. Skip research lab cosplay.

Inference APIs I run

Operate FastAPI /predict + Docker serve profiles — load artifacts, health checks. Ops around models, not inventing new ones.

FastAPI · Docker · Serving

Tabulate

Call OpenAI from Excel flows — keep people in the grid, not a side chat.

OpenAI · Spreadsheets · Integration

github.com/fudhoil/tabulate

Clear line

Applied work: use APIs, manage agents, ship product AI. Not training foundation models. Day job → HashMicro page.

Open HashMicro →

Applied AI stack

What I use to manage AI at work.

Model routing

Multi-provider APIs · OpenAI-compatible · Anthropic · token-lean context compression

Knowledge before code

HMX skill packs · personal skill hubs · provenance · pattern discovery

Orchestration

PM → Investigator → Executor → QA · worker desks · topic-routed channels

Tools & evidence

MCP (incl. OAuth) · PMT tickets/PRDs · search/fetch · tool-loop guardrails

Ship proof

Runtime validation · AI Score gates · Playwright/Cucumber E2E · full-auto MR

Ops surface

Messaging gateways · agent memory · sandboxed terminals · Cloudflare Pages

Elsewhere

Job and everything else, split out.

Current HashMicro Current job — use/manage AI + ERP delivery on International Team. Archive Other work Past roles, mobile products, open-source, education.

Contact

Need AI used well inside a product?

Agents, skills, MCP, or product intelligence — I wire and run it. Not looking to invent your foundation model.

fudhoilhlf@gmail.com GitHub LinkedIn