Bite v1 is Live

Last week we launched Bite v1. The response has been stronger than expected.

Two firms are actively using it. Several more are in the pipeline waiting to kick off. The pattern we're seeing confirms what we heard in dozens of discovery conversations: structural engineering firms are sitting on years of accumulated knowledge they can't access efficiently.

Every firm we've spoken to describes the same frustration. Project files scattered across servers. Past work buried in folder structures that made sense to whoever set them up five years ago. Senior engineers fielding the same questions repeatedly because they're the only ones who remember what the firm did on that hospital project in 2019.

The cost is significant. In a 20-person firm, we estimate 3,000 hours lost annually to searching for information. That's roughly £150k in productive capacity that could be spent on actual engineering work.

But the deeper cost is harder to quantify. Knowledge walks out the door when people leave. New hires take six to twelve months to learn where everything is. Firms miss opportunities because they don't realise they've solved similar problems before.

Engineers have tried using general-purpose AI tools to help. Upload a document to ChatGPT, ask questions, get answers. It works for simple cases but breaks down quickly in practice.

The first problem is context windows. Even documentation for a single project, the calculations, drawings register, specifications, correspondence, can exceed what an LLM can process in one conversation. You hit the limit and the model starts forgetting what you told it earlier. For company-wide knowledge spanning years of projects, context windows aren't even close to sufficient.

Then there's the tedium of the workaround. Copying and pasting sections manually. Attaching documents that don't parse properly. Reformatting PDFs that come through garbled. Engineers end up spending almost as much time preparing information for the AI as they would have spent searching manually.

The deeper issue is you don't know what you don't know. When you upload a specific document, you're making an assumption that the answer lives in that document. But what if the relevant precedent is in a different project folder? What if the company guidance you need is buried in an email thread from three years ago? What if there's a better approach documented somewhere you've never looked? Manual upload means you're limited to what you already know to look for.

Finally there's data security. Uploading project documents to external AI services raises serious questions about confidentiality. Client information, proprietary details, commercially sensitive data, all potentially leaving your control. For firms handling sensitive projects, this isn't a theoretical concern.

The concept is simple. We connect to your existing file server with read-only access. Our AI indexes your documents, design files, PDFs, Word files, spreadsheets, emails, and learns your firm's vocabulary and project history. Then anyone on your team can ask questions in plain English and get instant answers with citations linking back to source files.

"Show me similar projects we've done for education clients." Instant results.

"What was our approach to the foundation design on the Manchester warehouse?" Direct answer with the relevant documents surfaced.

"What's our standard detail for steel-to-concrete connections?" Found in seconds rather than hours.

The system watches for changes automatically. New files get indexed as they're added. Your knowledge base stays current without anyone maintaining it.

Our Technical Approach

Building AI that reliably answers questions about engineering projects requires more than a single method. Different types of questions demand different retrieval approaches, and getting this wrong means either missing information or returning confident nonsense.

Consider four questions an engineer might ask:

A single retrieval method can't handle all of these well. So we built Company Intelligence with multiple retrieval systems working together.

RAG for Contextual Questions

Retrieval Augmented Generation works by finding relevant document chunks and feeding them to the language model as context. It excels at semantic questions where the answer requires understanding and synthesis: design approaches, company guidance, lessons learned from past projects. RAG finds the relevant reports and lets the model explain what they contain.

But RAG has limits. If information isn't explicitly written somewhere, RAG can't find it. For large structured datasets like a complete element list, the information either doesn't exist in document form or gets truncated by context limits.

Structured Data for Deterministic Questions

For questions with specific, deterministic answers we expose structured databases directly to the model. When someone asks about beam B01's utilisation or wants a complete element list, the system queries the database and returns precise results. No interpretation, no synthesis, just the data.

This handles the questions RAG struggles with: specific values, complete lists, anything that lives in structured form rather than prose.

The Orchestrator

The challenge is knowing which approach to use for any given question. Our orchestrator pre-processes each query, determines whether it needs RAG, structured database access, or both, then routes accordingly.

This architecture matters for two reasons. First, Bite v1 can answer the full range of questions engineers actually ask rather than being limited to one retrieval pattern. Second, it's extensible. The orchestrator can incorporate additional capabilities over time, including the design agents we're developing for future releases.

Accuracy

Every answer includes citations to source files. The system only returns what it finds, whether from documents or structured data. When information isn't available, it says so rather than guessing.

This connects to our broader philosophy around AI in structural engineering. The technical architecture isn't just about capability, it's about building systems that engineers can trust and verify.

Structural engineering is knowledge work. The value firms provide comes from accumulated expertise, lessons learned across hundreds of projects, institutional memory about what works and what doesn't. But that knowledge is only valuable if people can access it.

We've watched firms grow and realise their knowledge systems don't scale. What worked when everyone knew each other and could shout across the office breaks down at 15, 20, 30 people. The coordination overhead starts eating into the time that should be spent on engineering.

Bite v1 is designed for this moment. It makes your firm's accumulated knowledge accessible to everyone, immediately, without changing how you store files or requiring new systems to learn.

Early Results

The firms using Bite v1 are reporting what we expected but it's good to see confirmed. Hours of searching compressed into seconds of asking. New team members accessing company knowledge from day one rather than spending months learning the filing system. Senior staff freed from answering repetitive questions about where things are.

One firm described it as finally being able to use the knowledge they'd spent years building up.

Data Security

Your files stay on your infrastructure. Bite v1 connects with read-only access and processes documents locally. No project data gets uploaded to external AI services. No client information leaves your control. For firms handling sensitive work, this is non-negotiable and we've built accordingly.

Bite v1 is the knowledge agent. It makes your firm’s accumulated expertise accessible instantly. But it’s the first and foundational agent of a team of agents coming soon.

Bite v1.5 ships in April with Excel calculation agents. Bite v2 ships in June with full 2D design agents covering Tedds and spreadsheet automation. Bite v2.5 ships in August with 3D agents for ETABS and SAFE.

Each version builds on the last. The knowledge agent informs the design agents. The agents act on the knowledge. By year end, Bite will handle the full loop from understanding your projects to executing design changes across your tools.

Bite v1 is available now. Pricing depends on firm size and locks in at the version when you sign. Firms who join at v1 keep v1 pricing until renewal. Firms who wait for v1.5 or v2 pay v1.5 or v2 pricing.

Early customers also get direct input on our roadmap and beta access to agents before general release.

The ROI calculation is straightforward. If it saves two hours per person per week, a 20-person firm recovers over 2,000 hours annually. That's a 10-20x return on the investment.

If you're running a structural engineering firm with years of project history sitting in servers that nobody can navigate efficiently, we should talk.

Book a walkthrough at bite.engineering or email [email protected].

Bite builds AI-powered workflow automation for structural engineers. We're working with partners across the UK to develop tools that save time without compromising safety.