Fintech
68%
tickets resolved without a human
Conversational support agent
A retrieval-grounded support agent wired into the client's ticketing stack, with confidence-based escalation to human agents.
- RAG
- Agents
- Evals
NewArticom is live nowWe design and ship
We design and shipAI agents that hold up in production.
Most AI projects stall at the demo. We take them the rest of the way: architecture, evaluation, guardrails and the engineering around them, so they survive real users, real data and real scale.
- AI agents & copilots
- Realtime voice AI
- Full-stack platforms
Teams we build with
What we do
AI first, and the engineering that makes it real.
Our centre of gravity is applied AI. But an agent is only as good as the platform around it, so we build that too.
Core
AI agents & copilots
Autonomous and human-in-the-loop agents that actually complete work: tool calling, memory, multi-step planning and a fallback path for everything that can go wrong.
- Tool & function calling
- RAG copilots
- Multi-agent orchestration
Core
Voice AI
Realtime speech agents that answer, qualify and resolve calls with sub-second latency across languages and telephony providers.
- Realtime speech-to-speech
- Telephony & SIP
- Multilingual
AI automation
Back-office and document workflows rebuilt around models: extraction, classification, routing and approval, with humans kept where they add value.
- Document intelligence
- Workflow automation
- Approvals & routing
Data & RAG platforms
The unglamorous layer that decides whether your AI is useful: ingestion, chunking, embeddings, hybrid retrieval and freshness you can trust.
- Ingestion pipelines
- Hybrid & vector search
- Knowledge bases
Web & product engineering
We build any kind of web platform, from customer portals and marketplaces to dashboards and internal tools, with or without AI in the loop.
- Next.js platforms
- Dashboards & portals
- APIs & integrations
Cloud, MLOps & evals
Deployment, evaluation harnesses, observability and cost control, so model changes stop being a leap of faith.
- Eval harnesses
- Tracing & observability
- Cost & latency tuning
How we work with AI
The gap between a demo and a system nobody has to babysit.
Anyone can get a convincing first result out of a model. The work that decides whether it survives contact with production is what we do.
01 · Discover
Feasibility before code
02 · Architect
Architecture that survives change
03 · Evaluate
Evaluation as a first-class citizen
04 · Harden
Guardrails and graceful failure
05 · Operate
Run it, measure it, lower the cost
Selected work
Systems in production, not proofs of concept.
A sample of what we've shipped across sectors. Details are anonymised where clients asked us to.
Hospitality
<700ms
median response latency
Voice ordering line
Realtime voice agent taking phone orders in three languages, integrated with the POS and dispatch system.
- Voice AI
- Telephony
- POS
Insurance
12×
faster first-pass triage
Claims triage pipeline
Document intelligence pipeline that reads mixed-format claim packets, extracts structured fields and routes by risk.
- Document AI
- Automation
Logistics
40k+
shipments handled monthly
Freight marketplace platform
A two-sided marketplace with live tracking, tender workflows and settlement, built end to end on Next.js.
- Next.js
- Platform
- Realtime
Healthcare
100%
answers carry source citations
Clinical knowledge assistant
A citation-first assistant over clinical guidelines, with strict grounding and a refusal path when evidence is thin.
- RAG
- Guardrails
- Compliance
Energy
9
legacy tools replaced
Field operations suite
Offline-capable web suite for field crews, with scheduling, asset history and an AI assistant over maintenance manuals.
- Web platform
- Offline
- Assistant
Reach
A small senior team, shipping across time zones.
We work remote-first from Colombo with clients across Europe, the Gulf, Asia-Pacific and North America.
0+
Projects delivered
0+
Clients partnered
0
Countries served
0+
Years building software
Our own products
The platform we built so AI teams stop rebuilding the plumbing.
Articom is our SaaS platform for building AI agents, voice agents and automations. Design an agent, connect your tools and data, ship it to chat or the phone line, then watch how it actually performs, without assembling the orchestration, evaluation and telemetry layer from scratch every time.
IWInforwaves
Good afternoon, team!
Here's what's happening with your AI platform today.
Active Users
128,942
Unique users across assistants
Message Activity
Report2.4M
Past 7 days
Total Calls
86,410
Past 7 days
Best practice
Optimize token usage
Use structured prompts and a concise knowledge base to lower monthly costs.
Total API Cost
$18.2K
Total Messages
9.6M
AI Assistants
146
Create AI assistants faster
Build, connect and launch AI assistants across every channel.
Token Usage
682K318K
Voice Minutes
412K88K
Token Usage Statistics
Last 7 days
DailyWeeklyMonthly
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Top Assistants
Support Agent
42.1KLive
Sales Copilot
28.4KLive
Onboarding Bot
15.9KLive
Recent Activity
New assistant deployed to production · 2m ago
Knowledge base synced · 1,204 docs · 18m ago
1M messages milestone reached · 1h ago
Agent builder
Compose tools, memory and routing visually or in code.
Voice agents
Realtime speech agents with telephony built in.
Automations
Event- and schedule-driven workflows across your stack.
Integrations
Connect CRMs, ticketing, databases and internal APIs.
Evals & analytics
Golden datasets, regression runs and live tracing.
Governance
Budgets, PII controls, audit trails and role-based access.
Articom is a separate product with its own site. Everything above is also available as a managed build from the Inforwaves team.
How we engage
A delivery process with no big reveal at the end.
Six stages, each with something concrete you can review. You always know where the project stands.
Discover
Discover
We map the workflow, the data and the accuracy bar, then prove the riskiest assumption on a slice of your real data.
- Feasibility report
- Success metrics
- Cost per task model
Architect
Architect
System design, model selection, integration contracts and a build plan you can hold us to.
- Architecture doc
- Integration contracts
- Delivery plan
Build
Build
Two-week increments, demoed live. You see working software continuously, not a big reveal at the end.
- Working increments
- CI & eval harness
- Staging environment
Evaluate
Evaluate
Golden datasets, regression evals on every change and human review loops until quality clears the agreed bar.
- Eval dashboards
- Quality sign-off
- Failure playbook
Launch
Launch
Staged rollout behind flags, observability wired up, on-call runbook handed to your team.
- Production rollout
- Runbook
- Team handover
Scale
Scale
Tune cost and latency, expand scope, and keep the system current as models and your business change.
- Cost optimisation
- Feature expansion
- Model upgrades
Tech we use
Boring, proven tools, chosen per problem, not per trend.
We keep the stack deliberately narrow so our team stays deep rather than broad.
AI & ML
- Claude
- OpenAI
- LangGraph
- LlamaIndex
- vLLM
- Whisper
- LiveKit
Data
- PostgreSQL
- pgvector
- Pinecone
- Redis
- Kafka
- ClickHouse
Product
- Next.js
- React
- TypeScript
- Tailwind CSS
- Node.js
- FastAPI
Platform
- AWS
- GCP
- Vercel
- Docker
- Kubernetes
- Terraform
What clients say
The feedback we hear most.
They were the first team that told us which parts of our AI plan would not work. That saved us a quarter and a lot of money.
Head of EngineeringFintech clientThe voice agent went live in nine weeks and handled our December peak without us adding a single seat to the call centre.
Operations DirectorHospitality clientWhat set them apart was the evaluation work. We can finally change a prompt without wondering what silently broke.
VP ProductInsurance client
They rebuilt a platform three vendors had failed to deliver, and shipped the first usable version in six weeks.
Chief Technology OfficerLogistics clientCareful, senior engineers who push back when the requirement is wrong. Rare, and worth it.
Director of DigitalHealthcare clientOur per-request AI cost dropped by more than half after their optimisation pass, with no drop in answer quality.
Platform LeadRetail client
Questions
Things people ask before we start.
If yours isn't here, send it over. We answer plainly.
Inforwaves is the engineering company: we design and build AI systems and web platforms for clients. Articom is our own SaaS product, a platform companies use to build AI agents, voice agents and automations themselves. Some client projects are built on Articom; many are not. Articom has its own site at articom.io.
Let's talk
Have an AI idea that keeps stalling at the prototype?
Tell us what you're trying to build. We'll tell you honestly whether it's worth building, and what it takes.
Colombo, Sri Lanka · working worldwide