AI as a platform, not a feature
Inference, retrieval, and agents delivered as APIs and product surfaces. Built in Chandigarh. Run by the team that owns the software.
ENDPOINT
POST https://api.opsiva.in/v1/chat
One contract for chat, tools, and structured output. Models can change. The product interface does not.
WHAT YOU GET
- Versioned models and a written deprecation window
- Usage, latency, and error rates on the same dashboard as the rest of the product
- A handover so your engineers can operate the layer without us in the loop
The layer
Six pieces. Use one, or wire them together as the AI platform under a product.
Chat · Completions
Inference APIs
Stable HTTP APIs for chat, structured output, and embeddings — versioned, metered, and ready for production traffic.
RAG
Retrieval
Ground answers in your documents, tickets, and product data so the model cites what it used.
Workflows
Agents & tools
Agents that call your APIs, keep memory, and finish a job inside the product — with a person still in the loop.
Domain models
Fine-tuning
Models trained on your corpus, evaluated against a golden set before they ever reach a customer.
Multimodal
Voice & vision
Speech, documents, and images as first-class inputs — not a bolt-on after the text path works.
Evals
Guardrails
Eval sets, red-team checks, and fallbacks so the system says it does not know instead of inventing an answer.
How it ships
01
Define the contract
The API, the data it may see, and the cases it must refuse are written down before a model is chosen.
02
Ground the model
Retrieval, tools, and evals sit in front of generation. The answer has to point at a source.
03
Ship behind a version
A stable endpoint, metering, and a rollback. The product does not depend on a playground.
Put AI behind an API
Tell us the product and the data. We reply within one business day.