Estimate AI
Construction, Estimating
An extraction platform that trains a dedicated model for each vendor, turning inconsistent construction estimate PDFs into structured, ERP-ready data.
AI & Automation
AI systems that do real work inside your business
We build AI systems that extract information from documents, handle tasks automatically and help your team find answers. They connect to the tools your business already uses, and we test every change to make sure they remain accurate and reliable.
Who this is forBusiness leaders who want AI to handle real work, reduce manual effort and deliver reliable results as their business grows.
Problems we solve
Common problems, and how we approach them
Problems we see often, and the systems we build to solve them.
01
The problem
Finance and operations teams manually entering information from invoices, estimates and other PDFs, because standard OCR tools struggle with real-world layouts.
How we solve it
We build AI systems that learn from your own documents and improve as corrections come in. Our self-improving invoice pipeline reached 96% extraction accuracy.
02
The problem
Customers asking the same booking, availability and account questions on WhatsApp, phone and chat, all day.
How we solve it
We build chat and voice agents that can check availability, make bookings and look up customer records using your existing systems. They use your business information to give relevant answers and are tested before they go live. Our founders built the assistant behind a diagnostics network's patient app, which has more than a million downloads.
03
The problem
Engineers piecing incidents together from logs, devices and dashboards, often in the middle of the night.
How we solve it
We build investigation agents that gather evidence in parallel, explain their findings with a confidence score and propose a fix for a person to approve. Our founders built the incident-investigation agents behind Nanites.
04
The problem
Knowledge spread across documents, websites and files, with no reliable way to get a straight answer.
How we solve it
We build AI assistants that search your own information to find relevant answers, show where those answers came from and let your team review conversations when needed. Our founders built the retrieval engine behind Booosted's chatbot builder.
Scope
What a project can include
The exact scope is agreed in discovery. Every build includes automated tests, deployment scripts and full access to the code.
Core build
Document extraction pipeline
Parsing, extraction and validation against a schema, with the output delivered into your ERP or database.
Core build
AI agent with MCP tools
An agent that works through typed tools on your own systems, with clear limits on what it is allowed to do.
Core build
Evaluation and monitoring
Evaluation sets built from labelled and synthetic cases, run on every change and in production, with tracing in tools such as LangSmith or Logfire.
Discovery
Feasibility and data review
An assessment of your data, the accuracy you need and the integration points, before a larger build.
Extended scope
Integration with existing systems
Connectors that move AI output into your ERP, CRM or database, with retries and error handling.
Related work history
Work in this area
Work our founders have delivered in previous roles, as independent engineers and for our own products, across AI, Salesforce and product engineering.
Nanites
Network operations, IT infrastructure
AI agents that investigate network incidents like a senior NOC engineer: parallel checks on real devices, correlated evidence with a confidence score, and a scoped fix for an engineer to approve.
Cococure
Hospitality, Nightlife
An AI concierge that handles restaurant and nightclub bookings over WhatsApp and voice: it checks availability, makes the booking and recommends venues through MCP server tools.
Self-improving invoice extraction pipeline
Finance operations, Procurement
An invoice extraction pipeline on a fine-tuned Gemini model that retrains every week on newly processed and corrected invoices, reaching 96% extraction accuracy.
Narratize
Product innovation, R&D
Primary AI development partner for Narratize's earliest generative features: document ingestion, technical briefs, research reports and scientific analysis.
Armor Wallet
Crypto, DeFi, Web3 infrastructure
An open-source MCP server and agent layer that lets AI assistants carry out real crypto operations: wallet management, swaps, DCA and limit orders, and staking.
Booosted
SaaS, Customer support
The retrieval engine behind a Finnish AI-chatbot product: businesses add their website or PDFs and get a grounded chatbot they can embed on their site.
Guided memorial-writing assistant
Funeral and memorial services
A guided writing product that helps families create a personal obituary: a gentle interview, then drafting through a self-correcting LangGraph workflow that scores and revises its own work.
AI assistant for a national diagnostics network
Healthcare, Diagnostics
An AI agent layer for a diagnostics network whose patient app has 1M+ downloads: plain-language lab-report explanations, test guidance, lab lookup and voice, orchestrated across MCP tool servers.
Doctor Notes
Digital health, Medical documentation
The AI pipeline behind a medical-visit recorder: one consultation recording becomes an accurate transcript, a plain-language visit summary, and clinician formats such as SOAP notes and referral letters.
Aurora
Market research, Consulting
An AI market-research platform with competitor analysis and a LangGraph-based TAM/SAM calculator, turning a founder's research process into self-serve tools.
Booking and follow-up agent for pet-service businesses
Pet services, Local business
An agent that works inside the channels pet parents already use: it answers Instagram and Facebook messages, qualifies the lead and books the appointment in MoeGo, keeping GoHighLevel up to date.
Parliamentary transcript refinement
Government, Public affairs
An AI application that turns raw transcriptions of Canadian Parliament sessions into clean final transcripts with accurate speaker labels.
ERP documentation from interface screenshots
ERP, Enterprise software
An AI workflow that reads ERP interface screenshots, recognises the screens, fields and workflows on them, and drafts user documentation.
Lead discovery and qualification for construction
Construction, Business development
An automated workflow that finds churches, schools and non-profits near a construction firm, spots the ones running fundraising or capital campaigns, and posts qualified leads to Microsoft Teams.
Private Q&A over investment notebooks
Financial services, Wealth management
A natural-language Q&A interface, built on Claude through Amazon Bedrock, over proprietary investment notebooks and financial content stored in S3.
HubSpot CRM automation with multi-source lead ingestion
Sales operations, CRM
A HubSpot CRM built from scratch. Leads flow in automatically from Product Hunt, X and web scrapers, the team works with it through ChatGPT and Claude, and n8n syncs WooCommerce orders.
Grounded knowledge assistant for Islamic education
Religious education, EdTech
A chatbot that answers questions about Islam from a curated library of religious texts, with signed-in users and conversation monitoring.
Thought-leadership content engine for a video platform
EdTech, B2B marketing
AI writing tools that take a B2B marketing team from idea to finished thought-leadership and outreach copy, plus a separate SDR agent for sales outreach.
Cost-aware multi-engine AI search for GEO
Search technology, Digital marketing
A custom search and synthesis engine that combines several search APIs, live scraping and routed LLMs to deliver GEO insights, such as how a brand appears in AI answers, at a lower cost per query.
Intelligent service experience
Enterprise customer service
A governed AI service architecture in which Agentforce works inside Service Cloud. It is grounded in Data 360 context and Knowledge, scoped by topics and instructions, and always able to hand off to a human agent.
Process
How we deliver it
Four stages, with working software to review at every milestone.
01
Data and success criteria
We review sample inputs, agree what good output looks like and how it will be measured, and set the privacy boundaries.
02
Build and fine-tune
We build the pipeline or agent, its prompts and tools, and fine-tune models on your data where that beats prompting alone.
03
Evaluate
Every iteration runs against evaluation sets of real and synthetic cases, checking accuracy, schema compliance and edge cases.
04
Launch and monitor
Deployment to your cloud with tracing and monitoring from day one, and a full handover of code and documentation.
Cost and timeline
What drives cost and timeline
Every project is priced from an agreed scope, milestone by milestone. These are the factors that determine its size:
- •How varied your inputs are, from clean digital PDFs to low-resolution scans and handwriting
- •Latency needs: real-time voice and chat, or batch processing
- •The number and age of the systems the AI must read from or write to
- •How much labelled data exists, and how much evaluation the use case needs
Every project includes
Fixed-price milestones, working software to review every two weeks, a 30-day warranty after launch, and full ownership of your code and IP as milestones are paid.
Request a fixed-scope proposalFAQs
Frequently asked questions
Will our data be used to train public AI models?
No. We use enterprise model endpoints, such as Vertex AI, Amazon Bedrock or Azure OpenAI, whose terms exclude your data from training, or self-hosted models where that is required.
How do you prevent hallucinations in business-critical processes?
We validate outputs against strict schemas, constrain which tools an agent can call through the Model Context Protocol (MCP), and run evaluation suites after every change and in production, so drift is caught before users see it.
Can the AI connect to our ERP or CRM?
Yes. Our founders have connected AI systems to Salesforce, HubSpot, GoHighLevel, MoeGo, payroll and accounting systems and custom databases, through REST APIs, webhooks and MCP servers.
Project brief
Plan your AI & Automation project
AI & Automation is preselected, and you can change it. A senior engineer reviews every brief and replies within one business day.
Step 1 of 3What you need
Prefer to talk? Book a 20-minute call ↗ or email contact@softbytes.co.uk