bryan tan
Menu
Systems [7] Outcomes [6] Capabilities [26] Experience [8y+] Contact
Let’s talk

Bryan Tan

AI & Automation Engineer

LLM agents, data pipelines and workflow automation — built alongside the people who use them, and kept running in production.

See the work

Systems running now

medcare
968 tests
chronos
541 tests
investing
nightly
finances
live
career vault
909 tests
assistant
live
pet content
approve, then build
melbourne, vic self-hosted
LinkedIn Contact

Everyone has something worth automating. These three were mine.

  1. Someone in the family gets sick.

    The bills and the lab reports start arriving and nobody has time to file them. A photo into a chat is the whole interface: a vision model reads the provider, the date and every line item out of it, and the dashboard is what the family looks at instead.

  2. Money that needs decisions you have no time to make.

    Two horizons are graded separately, and a bull and a bear take opposite sides of the same position before anything is recommended. The disagreement goes into the report rather than being smoothed out of it.

  3. Help that can act for you, and has to be stoppable.

    It reads the failing job and names the fault. Then "patch it" is refused at the tool layer — not declined by the model, unreachable — because that action was never granted to this profile.

/Systems

7 running
Walkthrough on request
$ pytest -q 968 passed medcare-bot active medcare-web active research-pull active tunnel active
medcare · today
2/53/5
2 of 5 doses taken3 of 5 doses taken
next: Vitamin D at 19:00next: none today
Iron supplement taken
Calcium 500 taken
Vitamin D 19:00taken
1415bills filed
3lab reports
1needs review

Family health records

PythonSQLiteLLM extraction
$ pytest -q 541 passed mode generate-only verifier graded cost cap enforced
chronos · queue
Waiting on you — 21
projecttaskstate
atlasretry flaky suite plannedrunningverified
ledgerbackfill 12 rows verified
1819tasks today
94%verifier pass
okunder cap

Autonomous task agent

PythonLLM orchestrationpytest
$ nightly-run horizons long · short debate bull / bear sizing real NAV delivery telegram
nightly report · book B
Δ vs hold +4.2%+4.6%
tickersignalgradesize
AAABUYB+3.1%
BBB WATCHBUY CB —2.2%
CCCHOLDA-5.4%

Investment monitoring and simulation

PythonMulti-agentIBKR
$ ingest --bank nab statements parsed categories llm + rules tax pack ready
tax pack · FY draft
Rental schedule
itemrouteamount
Repairsdeductible$482
Water ratesdeductible$196
New ovencapital$715
Agent feedeductible$318
Bathroom resealneeds a decision$940
23 line items need a decision gap flagged

Rental-property tax automation

PythonSQLiteDocument routing
$ pytest -q 909 passed claims verified overlay per-application output pdf + linkedin
career · board
NEEDS YOUR ACTION 45
Halcyon Systems — fit report ready 91%
Northwind Rail — pack has MISSING answers 92%
Cobalt Health — scanned 3d ago, not applied 88%
Meridian Data — draft sections unedited 74%
fit scored against 31 evidence bulletsresume claims re-checked against source-repo tests rescan 06:00

Resume and application pipeline

PythonTypstpytest
$ tool-guard policy allow-list on miss fail-closed providers 2 in chain profiles isolated
assistant · tool guard
nightly run failed again — can you look?
reading ledger/parse.py…
line 41: a merge left two copies of the same function, so imports break at collection.
patch it
tool blocked — not on this profile’s allow-list. Nothing was changed.
guard: fail-closed profile: personal

Personal assistant gateway

PythonMCPsystemd
$ weekplan propose cards awaiting approval $ sushi-build build approved slots only $ sushi-daily insights pulled doctor passing
pet content · last 30 days
38,41038,960views
612followers
1.8%shares/reach
platformpostsmedianeng.
instagram 2021 1,1405.1%
tiktok202407.4%
youtube18963.2%
2 posts1 post to classify queue ready

Pet content pipeline

PythonGenerative mediacron

Every interface shown on these cards is recreated with invented data — no production screenshot is published. Private systems; walkthrough and code review available on request.

/Measured outcomes

delivered in role, not in a side project
95% fewer

Human parsing errors cut across a 7-person data-annotation team

Asset Lead, CYP
1 week → 2 days

Manual query work against the asset data model, after local AI agents

Asset Handover Specialist
1 day → 10 min

Financial report assembly, once extraction ran in Power Automate

Asset Handover Specialist
3 days → 10 min

Team processing time per topic, after the handover workflow was rebuilt

Asset Lead, CYP
100%

Documentation compliance across 8 sites and multiple subcontractors

Completions Engineer
90% less

Rework, after asset coordinates were validated and extraction gaps closed

Completions Engineer

/Capabilities

26 skills · 7 systems

Everything on this page runs on the same habit: notes in plain Markdown, linked note to note, in one private vault, with a pipeline that turns whatever it is fed into a knowledge graph. This is that treatment applied to the page itself. Each skill is a node, joined to the systems above that exercise it; the rest were built up across the roles below.

Skills by area

AI & Automation: LLM agents, RAG / semantic retrieval, prompt engineering, Power Automate, Microsoft 365 Copilot, Copilot Studio, AI agent design, workflow design, process optimisation.

Data: SQL, Snowflake, Python, Power BI, data pipelines, data quality validation, data modelling, SQLite.

Platforms & Tools: Microsoft 365, SharePoint, QGIS, HP ALM, git, Linux, systemd, Telegram Bot API, Cloudflare.

What each system draws on

Family health records: Python, SQLite, prompt engineering, data quality validation, Telegram Bot API, Cloudflare, systemd.

Autonomous task agent: Python, LLM agents, AI agent design, workflow design, git, Linux.

Investment monitoring and simulation: Python, LLM agents, prompt engineering, data pipelines, SQLite, Telegram Bot API.

Rental-property tax automation: Python, SQLite, data pipelines, data modelling, prompt engineering, Telegram Bot API.

Resume and application pipeline: Python, data pipelines, data quality validation, Telegram Bot API, git.

Personal assistant gateway: Python, LLM agents, AI agent design, systemd, Telegram Bot API, Linux.

Pet content pipeline: Python, AI agent design, workflow design, systemd, Linux, data pipelines.

What each role drew on

Asset Handover, Metro Trains, 2025 — now: Power Automate, Microsoft 365, Microsoft 365 Copilot, Copilot Studio, Power BI, Snowflake, SQL, RAG / semantic retrieval, LLM agents.

Asset Lead · CYP, Metro Trains, 2023 — 25: SQL, Power BI, SharePoint, data pipelines, data modelling, data quality validation.

Completions, Metro Trains, 2023: QGIS, Power BI, data quality validation.

Verification, Leica Biosystems, 2022 — 23: HP ALM, data quality validation.

Quality Tech, Rinnai, 2018 — 21: process optimisation.

Drawn at build time from the same data file as every other number on this site. The vault itself is not pictured here, and no screenshot of it is published.

/Experience

8+ years
  • Developed and deployed custom local AI agents automating data reporting, semantic retrieval, and SQL query generation against the asset data model
  • Built a Microsoft Copilot Studio agent grounded on SharePoint document libraries, using markdown context files to link source documents into the agent's knowledge, that drafts SQL data-quality checks on request — demoed to the engineer who authors those checks
  • Engineered an end-to-end claims intake pipeline across SharePoint, Power Automate, Excel, Microsoft Teams and Power BI — a workbook dropped into a status-named SharePoint document library is validated on sheet name, the submitter is notified in Teams on either acceptance or rejection, the claim cells are extracted into a logging workbook and the Power BI report is refreshed, with the report data backed up weekly; used by my manager and team to pull claim amounts without opening a file
  • Extended the SQL agent to generate data-quality checks, validated by execution in Snowflake and submitted through the data owner's service-request process pre-formatted to their required submission format — approved without rework and promoted to the team's approved check set, replacing hand-authored validation SQL
  • Led a team of 7 on standardised data-annotation practices, improving downstream data integrity
  • Streamlined asset documentation and handover workflows, accelerating stakeholder approvals
  • Automated semantic-model database queries and data pipelines
  • Designed and implemented an EACR workflow and compliance framework from scratch under tight project deadlines
  • Validated asset coordinates in QGIS and resolved data extraction discrepancies
  • Standardised data collection with cross-company engineering teams and subcontractors
  • Executed end-to-end verification and validation for the BOND-PRIME staining platform ahead of EU market launch, ensuring IEC/ISO compliance
  • Tracked test execution, defect resolution, and requirement traceability in HP ALM across multiple product iterations

Education

Monash University Master of Advanced Engineering (Mechanical Engineering) · 2019-2020

Swinburne University of Technology Bachelor of Engineering (Mechanical Engineering) · 2014-2018

Languages

English native

Chinese native

Malay native

Spanish elementary

Have a process worth automating?

Melbourne-based, open to engineering roles and automation work. Send a note and I’ll reply with how I’d approach it.

Start a conversation