AI in the places it removes real work.
Not a chatbot bolted to your homepage. Assistants grounded in your own documents, extraction that ends manual data entry, automation for the process someone currently does by hand every Tuesday.
- 2 – 4 weeks
- scope to live
- Fixed price
- against a written scope
- Your data
- never used for training
Who this is for
You probably need this if…
01
Someone retypes documents all day
Invoices, delivery notes, forms, receipts. A person converts paper or PDFs into database rows, and the errors they occasionally make cost more than the hour.
02
Your team answers the same questions constantly
The answer exists — in a policy document, a manual, a folder somewhere. Finding it takes eleven minutes and interrupting a colleague takes two.
03
You've been sold AI that did nothing
A generic chatbot that knew nothing about your business, hallucinated confidently, and got switched off after a month. That experience is common and it is not your fault.
What we build
The work, in detail.
Assistants grounded in your data
Retrieval-augmented systems that answer from your documents and cite where the answer came from. When it doesn't know, it says so rather than inventing.
Document extraction
Invoices, POs, delivery notes, forms. Structured data out of unstructured paper, with a confidence score and a human review step for anything uncertain.
Workflow automation
Classification, routing, summarising and triage. The judgement calls that are too fuzzy for rules but too repetitive for a person.
Semantic search
Search that understands what someone meant rather than matching keywords. Works across documents, tickets, products and internal knowledge.
Vision & image processing
Receipt capture, condition assessment, quality checks, OCR. Anything where a photo currently gets looked at by a person and typed up afterwards.
Prediction & recommendation
Demand forecasting, reorder timing, churn signals, next-best-action. Built on your own history rather than a generic model.
Guardrails & evaluation
Test suites over real examples, measured accuracy, fallbacks when confidence is low. You get numbers on how well it works, not a demo and a hope.
Model selection & cost control
The right model for each task, caching, and routing cheap work to cheap models. AI features that don't quietly become your largest cloud line item.
Practice areas
The disciplines behind it.
You buy this as one engagement at one price. Underneath, it draws on 3 of our practice areas, and the same people cover all of them.
AI & Automation
AI applied where it removes real work, not where it reads well in a deck.
- Grounded assistants
- Document extraction
- Workflow automation
- Semantic search
Data & Sync
Getting the same number in two places, and keeping it there.
- Schema & modelling
- Delta sync
- Reconciliation
- Migration & backfill
Product Engineering
The core build. Web and mobile applications from scoped spec to production.
- Web applications
- iOS & Android
- API design
- Real-time features
How it runs
Four weeks, in order.
Week 1
Scope
We find the process worth automating, gather real examples, and define what 'working' means as a number. If the honest answer is that AI isn't the right tool, you find out here.
Week 2
Build
The pipeline, grounded in your data, with guardrails. Friday demo running against your real documents — not a curated happy path.
Week 3
Tune
Evaluation against a held-out test set, accuracy tuning, and the fallback behaviour for cases the model gets wrong or is unsure about.
Week 4
Ship
Integration into the software your team already uses, monitoring for drift and cost, and training on what the system can and cannot be trusted with.
What you get, concretely
- The feature live inside your existing software, not a separate tool
- Source code in a repository you own from the first commit
- An evaluation set and measured accuracy figures you can re-run
- Defined fallback behaviour for low-confidence cases
- Cost monitoring and per-request spend visibility
- Training on what to trust it with and what to review
- 30 days of post-launch tuning at no additional cost
Typical engagement
Fixed price · two to four weeks
A single extraction pipeline and a full assistant with retrieval are very different jobs, which is why we will not quote one before scoping it. Model usage is billed at cost and shown to you separately.
Technology
What we build it with.
Defaults, not requirements. If you already run something else and have a team who knows it, we work in yours.
- Models
- Claude, GPT, open-weight models where they fit
- Retrieval
- pgvector, embeddings, hybrid search, re-ranking
- Orchestration
- Structured outputs, tool use, evaluation harnesses
- Infrastructure
- NestJS, PostgreSQL, Redis, queued processing
Proof
We have built this before.
Not a reference we cannot name. Systems we designed, shipped and still operate, with the decisions written down.
Questions
What people ask before signing.
- Will our data be used to train models?
- No. We use enterprise API tiers where inputs are contractually excluded from training, and self-hosted models where the data genuinely cannot leave your infrastructure. This gets written into the contract rather than promised verbally.
- What if it gets things wrong?
- It will, sometimes — that's why evaluation is a named phase rather than an afterthought. We measure accuracy on a held-out set, set a confidence threshold, and route anything below it to a person. You see the real numbers before launch.
- Is this just a ChatGPT wrapper?
- For simple cases, using a good model well is most of the work, and pretending otherwise would be dishonest. The engineering is in retrieval, grounding, evaluation and the fallbacks — which is exactly the part generic chatbots skip, and why they get switched off.
- What does it cost to run?
- It depends almost entirely on volume, and it is usually small next to the build. We show you per-request costs during the build, set a spend alert before launch, and design for cheaper models wherever they perform just as well.
- Can you add this to software we already have?
- That's most of what we do here. We don't need to have built the original system, though we do need reasonable API access or database access to it.
- What if AI isn't the right answer?
- We'll say so in week one and you'll have paid for a discovery week rather than a build. Plenty of problems presented to us as AI problems are better solved with a query and a rule.
Related reading
Written on this, by us.
AI & Development · 8 min read
What AI Actually Changes About the Cost of Building Software
AI is a real lever on development cost, but not the one most vendors claim. Here is which parts of a build it moves, which it does not, and why.
AI & Development · 11 min read
Integrating an LLM into Your Product Without Regretting It
The two integration patterns that survive real users, what the API actually costs at volume, and the five failures that catch most teams in the first month.
AI & Development · 8 min read
AI Development Trends in 2025: What's Actually Working
Six things AI is genuinely good at inside production software, two that are still demos, and the test that separates them before you commit a budget.
What does someone on your team do by hand every week?
Describe the repetitive part. We'll tell you honestly whether AI is the right tool for it, and if it is, what it would cost.
Phone
+1 (407) 796-2376Reply time
One business day, from an engineer
Based in
Orlando, Florida · serving the United States
What happens next
- A reply within one business day, from an engineer
- A thirty-minute call, with no qualifying call before it
- A written scope and a fixed number, if it fits