AI that ships and saves hours — not demos.
Practical, production AI for Sri Lankan businesses — Claude & GPT apps, RAG-powered chatbots, document intelligence, workflow automation and custom agents. Built, evaluated and safely deployed.
Built for the shape of your business.
Three broad client shapes we design, build and operate for. Every engagement starts with the actual workflow — not a template.
Customer Support at Scale
RAG chatbots and AI-augmented agent desks in Sinhala, Tamil and English — with human-handoff, source citation and safety rails.
- RAG over your knowledge base
- Source citation + confidence
- Seamless human handoff
- Sinhala / Tamil / English
Document & Back-Office AI
Extracting structured data from invoices, KYC docs, lab reports, shipping manifests and legal contracts — with human review loops.
- Multi-page doc extraction
- Human-in-the-loop review
- Audit trails + explainability
- ERP / CRM writeback
Agents & Workflow Automation
Custom agents, Zapier / Make / n8n workflows and tool-using Claude agents that take real action across your stack.
- Custom tool-using agents
- Workflow orchestration
- Guardrails + approval steps
- Cost + latency monitoring
Everything a modern program actually needs.
No generic checklist — every capability is specified to the operational reality of Sri Lankan and international markets.
Claude & GPT Applications
Production apps on Anthropic Claude (Sonnet, Haiku) and OpenAI GPT — with proper eval suites, fallbacks and cost controls.
RAG & Knowledge Bots
Retrieval-augmented chatbots over your documents, website and internal wikis — with citation, freshness and access control.
Document Intelligence
Vision + text extraction on invoices, KYC, lab reports, contracts and manifests — with human-review loops for edge cases.
Agents & Tool-Use
Claude and GPT agents that call your APIs, query databases, send messages and complete real multi-step tasks with approval gates.
Workflow Automation
Zapier, Make, n8n and custom automation to replace manual data shuffling between CRMs, sheets, inboxes and support tools.
Evals, Guardrails & Safety
Every system ships with a golden eval set, hallucination detection, content filters, PII redaction and cost / latency dashboards.
Tangible outputs. No vapor.
Every engagement produces a specific set of assets, documentation and ongoing guarantees — handed over in a documented package.
Launch deliverables
- Use-case scoping + ROI model
- Data & access audit
- Working prototype (week 4)
- Eval suite + golden dataset
- Production deployment
- Guardrails + PII redaction
- Cost / latency dashboard
- Team training (3 hours)
Ongoing deliverables
- Monthly eval + drift report
- Model version upgrade management
- Prompt / retrieval tuning cycles
- Cost optimisation reviews
- Content expansion & re-indexing
- Incident response SLA
- Quarterly safety audit
- Annual architecture review
AI without evals is just vibes.
Every AI system we ship comes with a golden eval set, hallucination checks, PII redaction and cost / latency dashboards — so you know when the model is drifting, when a new version is actually better, and what every answered question costs.
Recent ai & automation engagements.
A sample of ai & automation engagements from the past three years — each measured against real business outcomes.
Sampath Bank Support
Trilingual RAG agent over 4,200 policy docs. -71% escalations, 38s average handle time down from 4m 12s.
Hemas Pharma Docs
Invoice + shipping-manifest extractor with human review. 92% auto-approval rate and 6.4 hrs / day saved.
Sri Lankan Airlines
Claude-powered baggage + booking agent. 1.8M queries / quarter with 4.6★ satisfaction and 22% lower support cost.
Pricing that matches the need.
Indicative engagement tiers. Every project is scoped to exact operational needs — request a proposal for a firm quote.
AI Pilot
- Single use-case pilot
- Claude or GPT integration
- Basic RAG over up to 500 docs
- Eval set (50 examples)
- Production deployment
- Cost + latency dashboard
- 30-day stabilisation
AI Platform
- 2–3 use cases integrated
- RAG + document extraction
- Agents with tool-use
- Human-in-the-loop flows
- Golden eval + guardrails
- Monthly tuning retainer (3mo)
- Team training programme
AI Transformation
- Multi-system AI platform
- Custom fine-tuning / adapters
- SSO · RBAC · audit logs
- Dedicated AI engineer
- Quarterly safety audits
- SLA-backed ops support
- Executive-level reporting
Answers to honest questions.
What clients usually want to know before engaging us on ai & automation.
Can AI actually save time in a Sri Lankan business — or is this hype?
Both are true. Hype cycles are real, but we've measured 30–80% time savings on specific back-office workflows (document extraction, customer triage, first-draft content). The trick is picking the right task: high volume, low creative ambiguity, with a clean eval. We start every engagement by scoping that first.
Do you use Claude, GPT, Gemini or something else?
All three, plus open-weights (Llama, Qwen) where privacy or cost matters. Model choice is a technical decision per use case — we benchmark on your actual data before committing, and design for model portability so you're never locked in.
What about privacy and PDPA compliance?
Every AI system ships with PII detection, redaction before model calls, configurable data-retention and full audit logs. For sensitive workloads we deploy self-hosted models or zero-retention API contracts. PDPA documentation is provided at handover.
Can the chatbot respond in Sinhala and Tamil?
Yes. We've shipped production chatbots in Sinhala, Tamil and English — with proper Unicode rendering, locale-aware formatting and retrieval that works across languages in your knowledge base.
How do we know it's actually accurate?
Every system includes a golden eval set (50–500 labelled examples), automated accuracy scoring on every model version, hallucination detection, and a live dashboard. If accuracy drops below threshold, we get paged before you do.
Who owns the data and prompts?
You — full IP transfer on prompts, code, evals and fine-tune adapters. API accounts (Anthropic, OpenAI) are in your name from day one. No vendor lock-in.
Services that pair naturally.
Most ai & automation engagements extend into adjacent work. Here's where projects usually grow.