§ 01Jeremy Tanasaleh

AI systemsthat survive production.

› engineering roles · contracts · scoped paid pilots

See selected work Explore paid pilotsroles · contracts · pilots

I turn brittle prototypes into reliable, human-governed software. Melbourne now; Auckland from March 2027.

Meridian 33 / corrected v2 verification

24 Jul 2026

verified locally

SOURCE
CLAIM
APPROVAL
DELIVERY

sources: 97
PDF pages: 27
adversarial checks: 29/29
tests passed: 146

Corrected-v2 evidence record. SMTP acceptance and sender sent-folder verification; recipient inbox receipt not claimed.

scroll ↓

pages in corrected v2

146

tests in corrected v2

29/29

adversarial checks passed

sources in its evidence corpus

§ 02

Selected work

Current systems I designed and built end to end.

PRJ-01

Meridian 33

private · human-gated

Creator / systems engineer · 2026

Private, local-first research engine that binds claims to dated sources, certifies HTML/PDF editions, and only delivers after an explicit human release gate.

stackPython · Pydantic · httpx · Jinja2 · Chromium PDF

evidence97-source release corpus · claim-to-source binding

qa146 tests · 29/29 adversarial checks · zero print overflow

delivery27-page corrected v2 · SMTP acceptance + sent-folder verification · immutable hashes

Case study →

PRJ-02

Tape

private

Solo developer

Observation-only equities-research terminals for US microcap biotech and the Indonesia Stock Exchange. Bloomberg-restrained, no buy/sell/target copy.

stackNext.js 16 · React 19 · FastAPI · DuckDB · Neon

datapoint-in-time ETL across SEC, FINRA, IDX and KSEI

rigorunit invariants, no-backfill rules and independent verification

craftWebGL/GLSL boot cinematic, two distinct design systems

Case study →

PRJ-03

Denise

live

Architect / lead developer · Kipanga, 2026

Multi-tenant AI email-triage and lead-handling engine. Onboarding a store is a config row, not a copied workflow.

stackn8n Cloud · OpenAI · Microsoft Graph · CoreBridge EVO

scaleconfig-as-data across franchise stores

rebuildchained-agent legacy design → shared two-tier engine

safetyschema-validated fail-closed I/O, escalated leads never auto-send

Case study →

PRJ-04

Kipanga AI Concierge

live

Lead developer · 2025–2026

Agentic chat widget driving inbound conversion: RAG over a Pinecone knowledge base with dual-model orchestration and guard-railed function-calling tools.

stackGPT-5.1 · Gemini Flash · Pinecone · Vercel cron

toolsmeetings, notification, ROI, retrieval and transcript workflows

infraSSE streaming, Turnstile, nightly hash-diff ingestion, CRM sync

Case study →kipanga.com &nearr;

PRJ-05

AI Outbound Engine

live

Lead developer · Kipanga, 2025–2026

Human-reviewed campaign operations engine that ingests configured lead signals, qualifies against an enterprise ICP, and drafts channel-specific outreach behind deterministic QA gates.

stackGemini 2.5 Flash/Pro · Apify · Serper · Pinecone RAG

engineresumable SQLite (WAL) state machine, YAML-config campaigns

qadeterministic gate: banned-phrase, AI-tell, flattery, hallucination

shippedlive campaigns delivered with human review

Case study →

PRJ-06

OnTheSet.ai

delivered

Sole developer · client: Kate Tiller

Film-production crew marketplace: a ground-up rebuild with a bespoke design-token system, an interactive Australia map, and a persistent call-sheet drawer.

stackNext.js 14 · TypeScript · Tailwind · shadcn/ui

ssgcrew profiles and location pages statically pre-rendered

searchtokenized search and hover-synced SVG map

Case study →

All work, in depth →

Also shipped

Enterprise Deep-Research AutomationDACH SAP CX consultancy · multi-stage source-to-report workflow

Bianca · screen-industry outbound agentNode.js · Pinecone RAG · Monday.com

NowActually · weekly HR notes pipelinePipedream · OpenAI · HubSpot · recurring synthesis

US private-aviation finance automationdual-entity QuickBooks · AOG claims · daily pipeline

SignsSA / Signarama · email + research agentGPT-4 · Pinecone · lead-response workflow

§ 03

How I work

The patterns I reuse across every system.

P01

Config as data

A new tenant or campaign is a config row, not a new workflow. Platforms scale by data, not by copies.

P02

Fail closed

Schema-validated LLM I/O contracts that reject before they spend. Escalated work never auto-sends.

P03

Verify, do not trust

Deterministic QA gates and independent passes catch false-green assertions before they ship.

P04

Own the whole lifecycle

First-principles architecture through deployment and observability, with the evidence and handover needed to operate the result.

§ 04

About

I design and ship production agentic systems, RAG applications, and LLM-powered automation end to end, from first-principles architecture through deployment and observability.

I am usually the sole or lead developer, owning the full lifecycle alone: reverse-engineering undocumented systems, instituting fail-closed safety and QA invariants, and turning brittle prototypes into config-as-data platforms that scale.

Alongside the AI work I build full-stack in Next.js and React, and develop independent equities-research tooling. University of Melbourne Computing and Software Systems graduate, open to product engineering roles and scoped consulting engagements.

Education

BSc, Computing & Software Systems, University of Melbourne

Computing and Software Systems major

Experience

AI Engineer & Automation Developer2025 to present

Kipanga

lead AI + automation across the client portfolio and internal products

Mobile Application Developer2025

Haven

cross-platform Flutter, AI features, performance tuning

AI Engineer & Web Developer2025

Dave Hunt International

RAG + prompt pipelines for a mental-health chatbot (chat-bestie.com)

Frontend Mobile Developer2024

No Moss Co

Auth0 + Flutter UI for the Pace It app (App Store)

Technology Consultant2023 to 2024

FSD/TSD, vendor-selection analysis, stakeholder decks (Jakarta)

§ 05

Stack

The tools I reach for.

AI / LLM / Agents

OpenAI, Claude, Gemini 2.5 · RAG · function calling · QA + safety gates · Pinecone

Languages

TypeScript, JavaScript, Python, Java, C/C++, Dart, SQL

Frameworks

Next.js 15/16, React 19, FastAPI, Tailwind v4, Motion, Three.js, GSAP, Flutter

Automation

n8n, Pipedream, Make.com, Relevance AI, Apify, Clay, Origami CRM

Data & Backend

Postgres/Neon, DuckDB, SQLite (WAL), Pinecone · point-in-time ETL · backtesting

Cloud & Infra

Vercel, Docker, k3s, Traefik, GitHub Actions, GHCR, Linux VPS

Integrations

Microsoft Graph, OAuth 2.0, QuickBooks, HubSpot, Monday.com, Resend, Twilio

Practice

config-as-data multi-tenancy · fail-closed contracts · idempotent writes · human-in-the-loop

§ 06

Contact

Open to AI, automation, and software roles, contracts, and scoped paid pilots

emailtanasalehjeremy@gmail.com&nearr;githubgithub.com/jst213&nearr;linkedinlinkedin.com/in/tanasaleh&nearr;instagraminstagram.com/jeremy_tanasaleh&nearr;

Email me

Colophon

§signedJeremy Tanasaleh

Built with Next.js 16, React 19, Tailwind v4, native scrolling, CSS motion and one optional WebGL shader. Type set in Fraunces and Geist. Source-controlled and continuously verifiable.