AVAILABLE FOR NEW ENGAGEMENTS · Q3 2026

We build the AI your enterprise actually needs.

Automaxis partners with enterprise teams to take AI from pilot to production. Agentic platforms, marketing analytics, and the data infrastructure underneath — measurable, defensible, and built to last.

Agentic platformsMarketing intelligenceData infrastructureAI search visibility
Email AutomaxisSee selected work
WHAT WE BUILD · 03 CAPABILITIES

Three things, done well.

FIG. 02 / CORE PRACTICE
01 / AI SYSTEMS
Agents that actually do the job.
Coordinator-subagent patterns. Scoped tool servers. Retrieval, evaluation harnesses, and citation enforcement. The unglamorous infrastructure that keeps an agent honest in production.
Claude Agent SDKMCPDeepSeek V4LangFuseEval harnesses
02 / ANALYTICS
Decisions, not dashboards.
Attribution, segmentation, churn, sentiment. Synthesized into one surface a non-technical operator can actually use — with the provenance to back every number.
pandasscikit-learnPlotlyStreamlitBigQuery
03 / INFRASTRUCTURE
The plumbing underneath.
FastAPI services, Pydantic contracts, Postgres with per-tenant RLS, ingestion pipelines. Hard isolation between the deterministic and the probabilistic layers.
FastAPIPydanticPostgres + RLSSupabaseGCP Cloud Run
SELECTED WORK / 03 SHIPPED

Three systems shipped
in the last six months.

HOVER A CARD — STACK REVEALS
01 · HEALTHCARE / INDONESIA
Hermina JKN Claims AI
A 5-module AI claims-processing system for one of Indonesia's largest hospital groups. Catches documentation gaps before BPJS submission, drafts evidence-cited physician queries, and gives 51 hospitals operational visibility into their claim pipeline.
51
Hospitals
5
AI modules
24-day
Build sprint
4
Parallel worktrees
Built on a two-plane architecture: a deterministic ingestion layer (~99.99% reliability, schema-validated) hard-isolated from a probabilistic AI layer with hallucination defenses — closed-world rule check, source-citation enforcement, no-new-diagnosis verifier. LLM proposes, deterministic rules dispose.
Stack
DeepSeek V4-FlashClaude Opus 4.7FastAPIPydanticPostgres + RLSNext.jsRechartsCerebras QwenEval harness
02 · AI ORCHESTRATION / 2026
Unified Marketing Intelligence
A multi-agent AI platform that lets a non-technical marketer ask a question in plain English — "which channel has the highest ROI?", "why did traffic drop?" — and get a synthesized, source-attributed answer drawn from attribution, customer, and sentiment data.
3
Subagents
12
MCP tools
11
API endpoints
5
Dashboard tabs
A coordinator agent decomposes the question, dispatches specialist subagents in parallel through 12 MCP tools, then synthesizes the findings into one source-cited answer. Same idea as a research team: manager, specialists, writer. Runs on realistic synthetic data simulating a full marketing ML stack.
Stack
Claude Agent SDKMCPFastAPIPydanticStreamlitPlotlypandasscikit-learnFaker
03 · ENTERPRISE BUILD / YELLOW PAGES SG
GEO + SEO Visibility Platform
A multi-tenant, white-label citation benchmarking platform built for Yellow Pages Singapore. They operate it, brand it as their own, and resell it as SaaS to their SME client base — Automaxis was the dev agency. Fixed-scope engagement, 8 weeks, 21 features, 30-day post-launch warranty.
5
LLM engines
50+
Query variants
21
Platform features
8‑week
Fixed-scope build
End-users submit a niche + region; Gemini expands it into 50+ semantic variants. A FastAPI orchestrator fans out concurrent requests across 5 LLM provider APIs with rate-limit-aware worker pools. Responses normalize into a unified citation schema; BigQuery aggregates historical runs for drift analysis; Claude generates a final report scored against the AICO framework with a 0–100 visibility index. Yellow Pages onboards their SME clients into branded tenant workspaces and bills via Stripe in SGD.
Stack
Next.js 14FastAPIasyncioSupabaseBigQueryAnthropicGoogle AIOpenAIStripe (SGD)LangfuseVercelGCP Cloud Run
HOW WE WORK / 04 STEPS

CFO-defensible.
No overclaim.

Olive AI and Watson Health failed because they over-promised on probabilistic systems. We anchor every number below published comparators, hard-isolate the deterministic from the probabilistic, and let the evals dictate when a feature is shippable.
01 / DIAGNOSE
Diagnose
A short engagement to map the actual problem. No tools picked yet. Usually 1 week, ends in a written scope.
02 / DESIGN
Design
Architecture, contracts, evaluation criteria. Where deterministic and probabilistic layers sit. Hard gates before any LLM call.
03 / SHIP
Ship
Build in 3–6 week sprints with hard module isolation. Nothing reaches a stakeholder demo until it clears the eval thresholds we agreed on at design.
04 / OPERATE
Operate
Observability via LangFuse. Cost ceilings. Nightly evals to catch model drift. Handoff with documentation, not vibes.
FOUNDER · AUTOMAXIS
Muhammad Fariz Ibrahim
AI, Analytics & Infrastructure Development
fariz@automaxis.xyzgithub.com/mufibraautomaxis.xyz
Most enterprise leaders know they should be doing something with AI. They just don't know where to start — and honestly, most of what's out there is hype.

I started Automaxis to help with that. Today the studio brings together specialists in AI/ML, agentic systems, software engineering and analytics — building agentic platforms, marketing analytics and the data infrastructure underneath for teams that want results without the technical headache.

We keep things simple, practical, and focused on what actually moves the needle.

If any of this resonates, send a note — we'll help you cut through the noise and figure out where AI is useful for your enterprise, and where it isn't.

CONTACT / 01 ENGAGEMENT

Tell us what you're
trying to build.

REPLY WITHIN 1 BUSINESS DAY

Skip the discovery-call dance. Send a short note about what you're trying to do and what's getting in the way. We'll reply with whether we can help and a rough shape of how.

Directfariz@automaxis.xyzgithub.com/mufibra
AVAILABILITY · Q3 2026
ENGAGEMENTS · 3–12 WEEKS
BASED · JAKARTA / SINGAPORE
REPLIES WITHIN 1 BUSINESS DAY