Lead AI Engineer

Hyde Park, MA · Hybrid · Full-time · $125,000–$150,000 + Benefits and Performance Bonus

We're looking for someone who wants to build something that doesn't exist yet. You'll lead the creation of an AI-powered operating system, one brain, across a family of businesses spanning collision repair, towing, auto repair, restaurants, real estate, property management and development. You'll take on real problems in accounting, dispatch, customer service, HR, marketing and operations, and help turn a 30-year-old, 150+-person company into a fully AI-native one. If complexity excites you and you want to see your work drive growth you can measure, this is the chance to build the operating system for an entire business ecosystem from the ground up.

Who we are

Akiki started as a single collision shop in Hyde Park in 1986. Today it's a family of companies with 155 people, a 60-truck towing fleet that runs about 5,000 jobs a month, and a collision operation certified by most major car brands. It's owner-led and independent. No investors, no committees.

We've grown by out-working everyone. Now we want to grow by out-thinking them.

The problem

Every Akiki business runs on its own software: collision estimating, shop management, towing dispatch, accounting, bill pay, point of sale, and so on. None of them talk to each other, so people are the glue. That works until it doesn't, and at our size and growth aspirations, it doesn't.

Your job is to build one brain across all of it. Every system feeds one central data layer, automatically. On top of that, AI agents for each function (accounting, dispatch, customer service, HR) watch their area around the clock, catch problems, chase them to resolution, and report to the owner every day.

What you'll do

You'll work directly with the owner & CEO and take the business one function at a time. For each function, you own the whole arc:

1.       Live it. Embed with the people doing the work. Ride in the trucks, sit with dispatch, shadow the accounting team. Learn how the work really happens, not how it's documented.

2.       Plan it. Write a complete engineering and business plan for that function: current state, data sources and their quality, target architecture, build-vs-buy decisions, success metrics, cost and expected return, and a rollout plan for the people whose jobs change.

3.       Build it. Ship the integrations, data models, agents and tools.

4.       Run it. Measure results against the plan, harden what works, cut what doesn't, then move to the next function.

Year one, roughly:

•         Foundation: one data layer fed by every core system, and a daily view of cash, payables and receivables across all companies.

•         Accounting: automated reconciliation and an accounting agent that drives every exception to resolution.

•         Dispatch: truck assignment and routing recommendations that know what each truck can and can't do.

After that: collision estimating built on manufacturer repair procedures, voice agents for insurers and customers, HR and time-and-attendance, and marketing.

The technical work

You are the entire engineering team, so this spans the whole stack:

•         Data ingestion. Reliable, monitored pipelines from accounting, bill pay, towing dispatch, shop management, estimating, telematics and bank feeds. Use APIs and webhooks where they exist; automate exports and extraction where they don't.

•         A shared data model. One model across legal entities: mapping charts of accounts, matching duplicate vendors, customers and vehicles across systems, and tagging transactions between companies.

•         Reconciliation engine. Deterministic matching first, LLM-assisted matching for ambiguous cases, confidence scores on every match, and an exception queue that people work from.

•         Agent architecture. Function-specific agents with scoped tool access (for example, MCP servers wrapping each internal system), triggered on schedules and events, with approval gates, audit logs and escalation by email, text or chat.

•         Optimization. Dispatch as a real-time vehicle routing problem with a mixed fleet and capability constraints (wrecker vs. flatbed vs. rotator, drivetrain, location, time windows), tested by simulation on historical call data.

•         Forecasting. Short-term cash flow forecasting by company, and call and job volume forecasting.

•         Document AI. Extracting data from invoices and statements, and retrieval over manufacturer repair procedures for estimating.

•         Evaluation and monitoring. Golden datasets, regression tests for every agent, and tracking of accuracy, drift and cost.

•         Infrastructure and security. Cloud deployment, containers, infrastructure as code, secrets, role-based access and backups. This system handles financial and employee data.

•         How you build. Agentic coding tools like Claude Code are your daily workflow. You direct AI to write code, then review, test and own what ships.

What you bring

•         1-3+ years shipping production software, including at least one year building LLM-powered systems that real users depend on.

•         Agentic systems experience. You've designed and shipped agents with tool calling, multi-step planning and execution, multi-agent orchestration and human-in-the-loop controls. You know how they fail and how to contain it. Experience with MCP or a similar tool protocol is a strong plus.

•         Fluency with agentic coding. Claude Code or a similar tool is how you work every day. You can break a large build into tasks an AI can execute, and you review, test and verify what it produces.

•         Expert Python and SQL. TypeScript is a plus.

•         Data engineering. APIs, webhooks, ELT pipelines, data modeling, Postgres and a cloud data warehouse. You can turn messy, inconsistent sources into a model people trust.

•         Algorithmic depth. You can formulate a messy business problem as something solvable: optimization, probabilistic record matching, time-series forecasting.

•         LLM evaluation. You build eval sets, regression tests and monitoring before you call something done.

•         Cloud and security fundamentals. Docker, infrastructure as code, access control and audit logging.

•         Business sense. You can write a plan a non-technical owner can approve: what it costs, what it returns, what could go wrong.

•         Ownership. You'll be the only engineer, at least at first. No one will hand you tickets.

Nice to have

•         Accounting or ERP data: general ledger, payables and receivables, reconciliation, multi-entity consolidation, QuickBooks or Sage Intacct APIs.

•         Vehicle routing, fleet telematics or GPS data.

•         Voice AI and telephony.

•         Document extraction and OCR.

•         Computer vision, for example dash-cam video.

•         A graduate degree in a quantitative field (CS, operations research, math, physics). Helpful, not required.

•         Any time around cars, trucks, shops or the trades.

This isn't for you if

•         You want a big team, a long roadmap handed to you, or a research job.

•         You don't want to get your boots dirty. You'll occasionally ride along with drivers and spend time on the shop floor.

•         Messy, real-world data frustrates more than it interests you.

How we work

•         Remote most days. On-site in Hyde Park at least once a week, plus a morning in the trucks or shop once a month.

•         You report to the owner and talk with him most days.

•         You'll have a senior AI advisor with 15 years of experience building AI systems for Fortune 500 as a sounding board.

•         You have budget for the tools and AI services you need.

What you get

•         $125,000–$150,000 base (depending on academic degree and years of experience), plus benefits and performance bonus

•         Your work shows up in the business within weeks, not quarters.

•         Real ownership of the technology strategy for a growing group of companies.

•         The chance to build an AI-native company from the inside, something most engineers only read about.

How to apply

Send us:

1.       Your CV

2.       Something you've built: a link, a repo, or a short write-up.

3.       A few sentences on the messiest real-world data problem you've solved, and how.