AI & Machine Learning
AI features that move revenue, not slide decks
Code Sage Labs builds production AI - chatbots, document Q&A, agents, and LLM features - for businesses in Sri Lanka and beyond. We use OpenAI, Anthropic Claude, and open-source models. We ship in weeks, not quarters.
Why AI now
AI is one of the few line items that pays for itself
Revenue impact
AI chatbots that qualify leads round the clock. AI search that lifts conversion. AI features that justify higher pricing tiers.
Time recovered
Automate the inbox triage, report generation, and document review that eats your team's week. Hours back, every day.
Production-grade
Evaluation harnesses, retry logic, prompt versioning, cost monitoring, and PII handling - the things that separate a demo from production.
Ship in weeks
First working version inside 2-4 weeks. We bias toward shipping, measuring, and iterating - not 6-month AI strategy decks.
Capabilities
What we build with AI
Seven concrete things we ship for clients. If your use case isn't here, send it over - most AI problems are variants of these patterns.
AI Chatbots & Assistants
WhatsApp Business and web chatbots that handle support, qualify leads, and close sales 24/7. Built on GPT-4, Claude, or open-source models.
Examples · Customer support automation, lead qualification, FAQ assistants
RAG & Document Q&A
Retrieval-augmented systems that let your team chat with your own data - PDFs, knowledge bases, contracts, manuals - with accurate, source-cited answers.
Examples · Internal knowledge search, contract analysis, policy Q&A
AI Agents & Automation
Agentic workflows that complete multi-step tasks across your systems - fetching data, calling APIs, making decisions, and writing back results.
Examples · Inbox triage, report generation, data enrichment, ticket routing
LLM Features in Your Software
Drop modern AI into your existing apps - smart search, autocomplete, summaries, classification, content generation - without rewriting your stack.
Examples · Email summaries, smart tagging, AI-powered writing tools
AI Search & Recommendations
Semantic search and personalized recommendations powered by embeddings and vector databases. Faster, smarter results than keyword search.
Examples · Product discovery, content recommendations, internal semantic search
Speech & Voice AI
Voice transcription, voice bots, and call analytics. Turn audio into structured data, automate phone support, or build voice-first products.
Examples · Call transcription, voice assistants, audio summarization
Custom ML & Fine-tuning
When off-the-shelf models aren't enough, we fine-tune open-source models on your data or train custom classifiers, predictors, and detectors.
Examples · Domain-specific classifiers, fine-tuned chat models, predictive scoring
AI stack
Tools we use in production
Foundation models
- OpenAI GPT-4 / 4o
- Anthropic Claude
- Llama 3
- Mistral
- Hugging Face
Frameworks
- LangChain
- LlamaIndex
- Vercel AI SDK
- PyTorch
- Transformers
Vector & data
- pgvector
- Pinecone
- Qdrant
- Weaviate
- PostgreSQL
Infra & deploy
- AWS Bedrock
- GCP Vertex AI
- Docker
- Kubernetes
- Modal
How we work
From idea to production in three steps
Discovery
A short paid discovery sprint. We map the use case, identify the right model and tools, and produce a fixed quote for the build.
Build
We ship a working prototype in 2-4 weeks. You see real progress weekly and can redirect scope as you learn what works.
Production
Evaluation, monitoring, cost controls, and integration with your existing systems. We hand over runbooks and stay on retainer if needed.
FAQ
Common questions
How much does an AI project cost in Sri Lanka?
Most AI integrations fall between USD 3,000 and USD 25,000 depending on scope. A WhatsApp chatbot or document Q&A system typically runs USD 3,000-8,000. A full custom AI agent or fine-tuned model integration runs USD 10,000-25,000. We start with a fixed-price discovery sprint so there are no surprises.
How long does it take to build an AI chatbot or AI feature?
A working WhatsApp or web chatbot ships in 2-4 weeks. RAG document Q&A systems take 3-6 weeks. Custom agents and fine-tuned models take 6-10 weeks. We deliver in weekly increments so you can use it and give feedback before the project ends.
Which AI models do you use - OpenAI, Anthropic Claude, or open-source?
All three. We pick based on your needs - OpenAI's GPT-4o for general tasks, Anthropic's Claude for long-context and reasoning, and open-source models (Llama 3, Mistral) when you need to run on your own infrastructure or fine-tune. We are model-agnostic and design systems so you can switch providers later.
Can your AI chatbots handle Sinhala or Tamil?
Yes. Modern models (GPT-4o, Claude 3.5, Gemini) handle Sinhala and Tamil reasonably well, and we test rigorously for your specific use case. For higher accuracy, we can fine-tune open-source models on Sinhala/Tamil data.
Do you build the AI from scratch or use existing models?
Both, depending on what makes sense for your use case. 90% of business problems are solved faster and cheaper by integrating proven models like GPT-4o or Claude. We only train custom models when there is a clear reason - domain-specific data, on-prem requirements, or cost at scale.
How do you handle data privacy and AI compliance?
We support multiple privacy postures: API-only with no data retention, self-hosted open-source models for sensitive data, and AWS Bedrock / GCP Vertex AI for compliance-friendly deployments. We follow standard practices around PII redaction, audit logging, and access control.
Can you add AI features to my existing software?
Yes - this is one of the most cost-effective ways to use AI. We add LLM-powered features (search, summaries, autocomplete, smart tagging) to your existing web or mobile app without rewriting your backend. Most LLM feature integrations take 1-3 weeks.
Do you offer ongoing support after the AI is live?
Yes. Most AI projects benefit from a retainer covering prompt tuning, evaluation, cost monitoring, and feature additions as you learn what works. Retainers start at USD 500 per month.
Got an AI use case?
30-minute call. We'll tell you if AI is the right answer, what it would cost, and how fast we can ship.