Skip to content
3rdLoopSolutions

AI that does the routine work. People keep the decisions.

Assistants, retrieval, document AI, and automated workflows. Each one ships with an owner, a risk level, and an escalation path, and is proven on one workflow before you scale it.

In scope: AI SolutionsIntelligent Automation

Start with a free assessment What it costs

01The engagement

Every phase ends with something you keep.

You see what each step produces before the next one starts, and you can stop at any milestone.

What you keep A point where you can stop
  1. 1.1

    Assess

    We map the workflow with the people who run it: each step, what it costs, and how much harm a mistake would do.

    You keepWorkflow map and risk level per step

    You can stop here and keep what this phase produced.
  2. 1.2

    Set the loop

    For each step we decide what runs on its own, what waits for a person, and what the system may never do.

    You keepOversight policy and a named owner

    You can stop here and keep what this phase produced.
  3. 1.3

    Prove it

    One workflow and one metric, live with your team. We measure against how the work runs today.

    You keepWorking pilot and a before‑and‑after report

    You can stop here and keep what this phase produced.
  4. 1.4

    Evaluate & harden

    We test the AI against real and edge cases, set cost caps, and rehearse failure: wrong answers, outages, rollback.

    You keepEvaluation report and go‑live checklist

    You can stop here and keep what this phase produced.
  5. 1.5

    Run & improve

    We watch escalations, overrides, and cost, and use every correction to make the next release better.

    You keepMonthly report on quality, oversight, and cost

02Scope

What we deliver.

Software that does routine work and escalates the rest, with an owner, a risk level, and an escalation path from day one.

Call us when

  • A frequent workflow takes hours of skilled time every week.
  • Answers live in documents that are hard to search and harder to cite.
  • You tried a generic AI tool and couldn’t trust or audit what it did.
  • A regulator, board, or client will ask who approved each action.
  1. 2.1

    AI Solutions

    Assistants, retrieval, and document AI with evaluation and oversight built in.

    • Assistants grounded in your documents, with every answer cited
    • Retrieval over contracts, policies, records, and knowledge bases
    • Document extraction, classification, and drafting
    • An evaluation set that scores the AI before each release
  2. 2.2

    Intelligent Automation

    Workflows that run routine work and escalate exceptions to a named owner.

    • Workflows that run across your software, documents, and people
    • Review and approval queues for the steps that carry risk
    • Escalation to a named owner, with the full context kept
    • An audit log of every action and the evidence behind it

03Fees

Quoted before the work starts.

Solutions are project work: quoted after the free assessment, and billed apart from any product subscription.

  1. 3.1

    AI Solutions

    Proof of value, then scoped build

    We prove it on one workflow and one metric first, as a fixed‑fee pilot of 4–6 weeks credited against the build. The build is quoted after that, with AI usage passed through at cost or capped by an allowance.

  2. 3.2

    Intelligent Automation

    Per workflow

    Each automated workflow is scoped and quoted on its own, so you can add the next one only when the last has paid off.

3.3

Prove one workflow before you buy the build.

After the free assessment, we run one workflow live with your team for a fixed fee, against a result we agree in writing first.

Fixed fee
₱350k–₱750k
Term
4–6 weeks
How it’s set
Fixed fee, set by the workflow’s size and integrations. The full fee is credited against the build if you continue.

Where your pilot sits in the range

  • How many systems it connects to
  • Document and case volume
  • How ready the data is
  • How many steps need human review

Agreed in writing before it starts

  1. (a)The metricHours saved, cycle time, or error rate: one number, measured the same way before and after.
  2. (b)The test sampleThe real cases the pilot is measured on, chosen with your team.
  3. (c)The pass markThe result the pilot must reach before anyone talks about a larger build.

You keep the findings, the oversight policy, and the report, whether or not you go on to the build.

What the fee pays for

  • Assessment into baseline

    Mapping the workflow with your team and measuring how it runs today, so the result has something to beat.

  • Setting the loop

    An owner, a risk level, and an escalation path for every step, written up as your oversight policy.

  • Building the pilot

    The working workflow, connected to the systems and documents it needs, live with your team.

  • Testing and evaluation

    Runs against the agreed sample, including edge cases, with every error and escalation logged.

  • AI usage and hosting

    Model usage, hosting, and monitoring for the length of the pilot, with no separate bill.

  • Results

    A results demo and a before‑and‑after report against the agreed metric.

04Tools

Tools we run in our own products.

The same stack behind Batayan, ariarian.ai, and Habi. We pick per project, and work with yours where it already exists.

  1. 4.1

    Models

    Chosen per task on accuracy, cost, and where your data may go. Swappable without a rewrite.

    Anthropic Claude, OpenAI, Google Gemini, Open models via Ollama

  2. 4.2

    Agents & orchestration

    Typed tools and structured outputs, so the AI can only take the actions we allow.

    Pydantic AI, Laravel AI, FastAPI

  3. 4.3

    Retrieval & search

    Answers come from your sources, and link back to them.

    PostgreSQL + pgvector, Meilisearch, PyMuPDF, unpdf

  4. 4.4

    Collection

    Crawlers that keep public and regulatory sources current.

    Playwright, Crawl4AI, Beautiful Soup

  5. 4.5

    Workflows & queues

    Long jobs run in the background, retry safely, and never block a person.

    RabbitMQ, Redis, Background workers

  6. 4.6

    Evaluation

    A test suite for the AI itself, run before every release.

    pytest evaluation sets, Regression fixtures, Cost and latency tracking

05Standards

What doesn’t change from project to project.

  1. 5.1

    Every answer shows its source

    Retrieval answers link to the document, clause, or record they came from, and say when the evidence runs out.

  2. 5.2

    Measured before it ships

    We agree a pass mark with you. A release that scores below it on the evaluation set doesn’t go live.

  3. 5.3

    Costs you can cap

    AI usage is tracked by team and workflow, with limits you set. No surprise bills.

  4. 5.4

    Your data, your choice of model

    We route to the provider and region your policies allow, or run open models on infrastructure you control.

06Questions

AI & automation, answered.

Which AI model do you use?

The one that fits the task. We build on Claude, OpenAI, Gemini, and open models, and pick per workflow on accuracy, cost, and data rules. Because the model sits behind one interface, we can switch later without rebuilding.

Can it run without sending our data to an outside AI provider?

Yes. Where your policies require it, we run open models on infrastructure you control, such as your AWS account or a private VPS, or route only to providers and regions you approve.

What happens when the AI gets it wrong?

The oversight policy decides. Low‑risk steps are logged and sampled, medium‑risk steps have a review window, and high‑risk steps wait for a named person. Every correction is recorded and feeds the next evaluation.

How do you know it works?

Before the pilot we agree one metric, such as hours saved, cycle time, or error rate, and measure it against how the work runs today. The evaluation set checks quality before every release.

Do we need clean data first?

No. The assessment shows which sources the workflow needs and how usable they are. Where data needs work, we scope that as part of the build.

Start with a free assessment

The rest of how we work.

Your people stop doing the routine work. They still make the calls that matter, and they can always step in.

Tell us about one workflow that eats time or carries risk. We’ll reply with what software could handle, what it should hand to a person, and who stays accountable.

Get a free workflow assessment