Aufstin Filiko AI and automation engineer

I build the AI systems companies actually run on.

Agents, voice agents, data platforms and the automation underneath them, for founders and small teams who need it working in production, not just in a demo.

Start a project See the work
  • $1M+wager volume in the WagerVS beta
  • 95%+PHI detection confidence for AHDX.org
  • 8–15sGistify.co responses, down from 20–30s
  • 95%less manual data entry at SafeWay

01What I build

Five kinds of work, one standard: it has to run.

AI agents and automation

Agents that do real work inside your tools: scheduled runs, browser automation, triage bots and internal tooling. I run a platform of about 45 AI tools across lab, quality, operations, finance and commercial teams.

Data platforms and agent-readable data layers

One data layer that dashboards and agents both query, served through an MCP server, with provenance so every answer traces to its source record. Schema design and migrations included.

Voice agents

Inbound and outbound phone agents on Vapi, Twilio, ElevenLabs and Cartesia, built for regulated work. One system I built handles hundreds of financial calls a day.

Full-stack MVPs, web and mobile

Next.js and React on the web, React Native and Expo on mobile, Node or FastAPI and PostgreSQL behind them. Past builds include a Lightning wallet, a translator for a university and storefronts with live POS sync.

Agentic QA and testing

Agents that drive your product end to end the way a user would. On a Bitcoin Lightning wallet, agents running real payment flows against real Lightning nodes replaced manual QA.

Something else in mind?

If it involves AI, data or automation, describe it. If I am not the right fit, I will say so.

Tell me about it

02Selected work

Shipped for startups, schools and small businesses.

Independent contract work from October 2024, across fintech, gaming, healthcare, education and e-commerce.

WagerVS

wagervs.fun

Peer-to-peer prediction platform for streamers.

  • An AI wager generator: batches of 1 to 20 wagers with generated text and images from a plain-language prompt, with category targeting, bulk approve or reject, and JSON export.
  • The AI-oracle settlement architecture: TEE-signed model verdicts with confidence scores, and on-chain escrow with atomic payout, so a contested result is answered with a signed audit record.
  • A production migration from SQLite to PostgreSQL with a redesigned schema, which doubled query speed.
  • LLM pipelines
  • PostgreSQL
  • Solidity escrow
  • TEE-signed verdicts
  • 0G

$1M+in wager volume during the beta

2xquery speed after the PostgreSQL migration

Accepted into the Solana Foundation x OnePiece Labs accelerator and the 0G incubator.

AHDX.org

ahdx.org

HIPAA healthcare data platform.

95%+confidence on PHI detection

  • PHI detection with LangChain and locally hosted Ollama models, with heuristic pre-checks first to keep API costs down.
  • Data routed across AWS S3, Filecoin and Azure Blob by AI-assigned classification, with full audit logging.
  • LangChain
  • Ollama
  • React
  • Node.js
  • PostgreSQL
  • S3, Azure Blob, Filecoin

Gistify.co

gistify.co

Multi-LLM platform that sends each query to the right provider.

20–30s 8–15sresponse latency, at under $0.01 per query classification

  • Query classification in Python that routes every request to the best provider for it.
  • Parallel processing and real-time response streaming across the major LLM APIs.
  • Python
  • LLM routing
  • Streaming

SafeWay Driving School

Paperwork automation for a Houston driving school.

95%less manual data entry

90%less verification time

  • A Python pipeline that generates and analyzes student driving logs from PDFs.
  • Image analysis and OCR that parse handwritten and scanned documents, plus interface changes so staff adopted it day to day.
  • Python
  • OCR
  • PDF processing

Voice agents for a financial firm

Phone automation under strict compliance requirements.

Hundredsof regulated inbound and outbound calls a day

  • A voice-agent system that handles the firm's inbound and outbound financial calls.
  • Vapi
  • Twilio
  • ElevenLabs
  • Cartesia

A Bitcoin Lightning wallet

A production mobile wallet and backend for Lightning payments.

  • The full system design: a React Native (Expo) app on a Node.js and Express backend, with PostgreSQL on AWS.
  • Agentic QA that runs end-to-end payment flows against real Lightning nodes, replacing manual QA across the wallet-to-backend pipeline.
  • React Native
  • Expo
  • Node.js
  • PostgreSQL
  • AWS
  • Lightning

SmokeDashConnect

Storefronts for retailers, wired to their point of sale.

  • Configurable Next.js storefront templates with live POS sync for Square, Clover and Revel.
  • AWS hooks that keep inventory and catalog data in sync across storefronts, plus agentic workflows that automate UI testing.
  • Next.js
  • Square, Clover, Revel
  • AWS
  • MCP tooling

Universidad San Francisco de Quito

A Kichwa-Spanish translator for indigenous-language education.

  • A full-stack translator, with HuggingFace models for language understanding and LLM-driven practice conversations that give real-time feedback.
  • FastAPI services for translation workflows and structured vocabulary and grammar rules.
  • Next.js
  • React
  • FastAPI
  • PostgreSQL
  • HuggingFace

03Case study

Nexus Agriscience

AI & Automation Lead, since May 2026. My current full-time role.

The problem

A decade of legacy genetics records whose identifiers disagreed across the old stores, and lab and quality work spread across disconnected systems.

“Molecular farming: flavor, fragrance and functional ingredients.”

How Nexus describes itself

I am the architect on the company’s seed-to-sale platform, covering genetics, cultivation, refinement, the QA lab, quality and sales, and I run its internal AI tooling. In biology the data underneath has to be right before any model on top of it can be, so that is where most of the work starts.

What I built

Platform

Seed-to-sale platform architecture

One TypeScript system from genetics to sales, on a canonical data architecture I own: a unified ERD and phased surrogate-key migrations across the seed-to-sale schema.

~125API routes

  • Genetics
  • Cultivation
  • Refinement
  • QA lab
  • Quality
  • Sales

Data Atlas

One data layer, behind an MCP server

Every source system lands in cached, auto-refreshing record sets that agents and dashboards query through one MCP server and search API. A provenance graph means any answer traces back to its source record.

246controlled documents made fully searchable

PhyTree

Strain lineage as a living phylogeny

I migrated a decade of legacy genetics records (2013 to 2022) into the rebuilt platform. Identifiers disagreed across the old stores, so a pattern-detection pass rebuilt the lineage with confidence scores and gave researchers about 300 parent-child pairs to confirm instead of about 30,000 raw records. PhyTree renders the result as interactive phylogeny trees.

~300parent-child pairs to confirm, instead of about 30,000 raw records

Illustrative Generated tree, not company data

Quality

A six-dimension batch-release gate

Release approvals run as a transactional outbox with event-driven fan-out, so a vendor outage becomes a retry instead of a half-completed product release.

Lunch & Learn

AI training for the whole company

I ran the program department by department, wrote 12 project-knowledge briefs, and set up recurring agent tasks so shared context outlives any single chat.

12project-knowledge briefs

Internal platform

An internal platform of about 45 AI tools

Agent schedulers in isolated git worktrees, browser automation and deploy-triage bots, automating work across lab, quality, operations, finance and commercial teams. For the analytical lab that includes GC-FID run planning.

Illustrative Synthetic trace, not instrument data. Peak positions are schematic.

04How it works

A clear scope, then working software early.

  1. Discovery call

    We talk through the problem, who it affects and what done looks like. You leave with a straight answer on whether I am the right fit.

  2. Scoped, fixed proposal

    A written scope with deliverables and milestones, and a fixed quote. Every quote is scoped to the project, so you know the cost before work starts.

  3. Short, shipped increments

    Working software early and often, deployed where you can use it. No big reveal at the end, and no surprises in between.

  4. Handoff with docs

    Documentation your team can follow and a walkthrough of how it all fits together, so the system keeps running without me.

05About

Hi, I'm Aufstin.

I am an AI and automation engineer. From October 2024 I built AI systems as an independent contractor for startups and small companies across fintech, gaming, healthcare and e-commerce, and since May 2026 I have led AI and automation at Nexus Agriscience.

Before that I was an automation engineer at Corteva, where I built a computer-vision model that estimates plant weight from camera images, wired imaging stations into Ignition SCADA, and designed a robotic laser-cutter fix projected to recover $450,000 a year in lost runs. I studied Biotechnology and Computer Science at the University of Houston, so labs, regulated data and real operations feel familiar. I work fluently in English and Spanish.

Education
B.S. Biotechnology and Computer Science, University of Houston
Languages
English and Spanish, fluent
Works with
Python, TypeScript, Node.js, React and Next.js, PostgreSQL, Docker, AWS, LangChain, Ollama, MCP

06Contact

Tell me what you want to build.

A few lines are enough to start. I reply by email, and if it looks like a fit we set up a discovery call.

Prefer email? ajfiliko@gmail.com

Your range helps me suggest the right scope. Every quote is scoped to the project.

The problem, who it is for, and anything already in place.

Your details are used only to reply to you.