Services
Ten things we build
Every one of these started as something we needed ourselves. That's why we can tell you what it takes, what it costs, and where it breaks.
01
Voice agents
Real-time voice that answers the phone, holds a conversation and takes the action at the end of it.
- Telephony-grade: inbound, outbound, transfer and escalation
- Low-latency pipelines, tuned for interruption and accents
- On-device speech where latency or ownership rules out the cloud
02
Knowledge-gated assistants
Assistants that answer strictly from your corpus — and refuse when the corpus has no answer.
- Retrieval over documents, tickets, contracts and recordings
- Refusal behaviour you can test, not just hope for
- Citations back to the source passage
03
Lead & revenue intelligence
Find the right accounts, understand them, and open the conversation with something worth reading.
- ICP discovery and enrichment from the open web
- Fit scoring with the evidence attached
- Outreach drafting in your voice, reviewed by a human
04
Internal tools & ops consoles
The tool your operations team keeps asking for, instead of the spreadsheet-and-email workaround.
- Approvals, queues, admin panels and audit trails
- Replaces manual handoffs with typed, testable code
- Built to be handed over, not rented back to you
05
Multi-agent orchestration
Several agents, clear roles, one outcome — with streaming progress a human can watch and stop.
- Tool-using agents with scoped permissions
- Streaming UX over SSE, with cancel and retry
- Frameworks chosen for the job (AgentScope, ADK, custom)
06
Document & data pipelines
Messy inputs in, structured verified outputs out — the unglamorous layer every AI product needs.
- Ingestion, OCR, parsing and normalisation
- Schema-validated extraction with confidence scoring
- Reconciliation against the system of record
07
Support & helpdesk copilots
Triage, draft, and resolve — with the agent doing the reading and your team doing the deciding.
- Ticket classification and routing
- Draft replies grounded in past resolutions
- Knowledge-base gaps fed back to the owners
08
Private & on-prem AI
The same capability with none of the data leaving your building — for regulated and cautious buyers.
- Self-hosted inference on your hardware or private cloud
- Data residency and access controls by design
- Cost modelling: self-host versus API spend, per workload
09
Agent observability & evals
The part that keeps it working after launch: what it did, what it cost, and whether it got worse.
- Tracing of every step, tool call and token
- Latency and cost dashboards per workflow
- Regression suites that run on every prompt change
10
Integration & modernisation
The systems you already run, finally talking to each other — and to the new agents.
- API glue, webhooks and event pipelines
- Typed backends in Go, TypeScript, Python and Rust
- Auth hardening and legacy surface reduction
Engagements
How a project usually runs
Four phases. You can stop after any of them, and you keep everything produced up to that point.
01
Diagnose
We find where the real time and money go, and whether an agent is the right answer at all.
02
Prototype
A narrow working system against your data, fast — so you judge capability, not slides.
03
Harden
Auth, evals, observability, failure modes and cost limits. This is where demos become products.
04
Run
Deployed, monitored and documented, with your team able to operate it and us still on the line.
Not sure which one is yours?
Describe the process that's costing you the most time. We'll tell you where it lands — and if it doesn't land anywhere, we'll say so.