dhī · Sanskrit for intellect, insight
AI that thinks before it acts.
Dhii is an independent AI engineering studio. We build agentic automations, AI agents and knowledge assistants, and we fine-tune, train and self-host models — proven on your data before you commit, and run with humans approving what matters.
Mumbai-based · working with teams worldwide · fixed written quotes
Built & operated by our team
OptyCV · AI career platformairis OS · AI workspaceedminOS · School management SaaSDhii Chat · Self-hosted AI chatWhat we build
Seven practices, one engineering discipline
Agentic AI & Automation
Agents and workflows that take actions across your tools, with humans approving what matters.
Assistants, Chat & Voice
Grounded assistants that answer from your knowledge on web, WhatsApp, Slack, Teams and phone.
Model Training & ML
Fine-tuned LLMs and custom models trained on your data for vision, language, forecasting and ranking.
AI Infrastructure & MLOps
Private model hosting, evaluation, guardrails, data pipelines and cost control for AI in production.
AI Product Engineering
AI features inside your product, new AI SaaS MVPs, and document intelligence.
Strategy, Security & Training
Readiness audits, compliance-aware design and hands-on training for your team.
Function-Specific AI
Packaged solutions for recruitment, education, e-commerce and support teams.
Have something that doesn't fit a box?
Describe the problemThe Dhii method
Insight first. Then action. Then autonomy — earned.
- 01
Map
We sit with the people who do the work and document the real process — every decision, exception and system. Each step is marked automate, assist or keep human.
Workflow map & autonomy plan
- 02
Prove
A narrow prototype runs on a few hundred of your real cases. We report accuracy, escalation rate and cost per task — and say so plainly if it isn't worth building.
Measured feasibility & fixed quote
- 03
Build
Production engineering against an agreed spec: typed tools, guardrails, integrations, evaluation suite and tracing. Weekly demos on staging.
Working system + test suite
- 04
Shadow
The system runs alongside your team, proposing rather than acting, until the numbers show which categories are ready for autonomy.
Go-live evidence by category
- 05
Operate
Autonomy is switched on step by step with monitoring, cost reports and drift alerts — on a retainer, or handed to your team with a runbook.
Runbook, dashboards, handover
Why teams choose Dhii
Six commitments
Evidence before promises
Every engagement begins with a prototype on your real data. You see accuracy and cost per task before you commit to a build.
Humans approve what matters
Consequential actions go through approval thresholds you control. Autonomy is earned per category, with the numbers to back it.
Model-agnostic, no lock-in
We pick models on your evaluation set — frontier APIs, open-weight or self-hosted — behind an abstraction so you can switch later.
You own everything
Code, prompts, datasets, model weights and infrastructure accounts are yours from day one. We resell nothing and take no vendor commissions.
Private when it needs to be
Sensitive workloads can run on open-weight models on your own servers, so prompts and documents never leave your environment.
Fixed, written quotes
After discovery you get a fixed price against a written spec — not an open-ended hourly meter.
Work
Products we build and run ourselves
AI career platform
OptyCV
A career-intelligence product that scores job openings against a candidate profile, tailors CVs and drafts outreach. Built and operated in-house.
- Python
- Celery
- PostgreSQL
- Redis
- Playwright
AI workspace
airis OS
A local-first desktop workspace that turns a written description into finished work — decks, documents, interface screens, landing pages, campaign assets, images and video — exported as real files.
- LLM orchestration
- Desktop app
- Generative media
School management SaaS
edminOS
A multi-school management system covering students, teachers, academics, timetables, attendance, exams, lessons, assignments and announcements, with self-serve school registration.
- SaaS
- Multi-tenant
- Web
Self-hosted AI chat
Dhii Chat
A private, self-hosted AI chat workspace running Open WebUI on top of locally served open-weight models via Ollama — no prompts leave the server.
- Open WebUI
- Ollama
- Docker
- Nginx
- Linux VPS
Insights
Field notes
-
2 Nov 2026 · 6 min
What Is Agentic AI Automation? A Practical Guide for Business Teams
Agentic AI automation explained without the hype — how AI agents differ from chatbots and RPA, where they work well, where they fail, and how to start safely.
-
26 Oct 2026 · 6 min
Designing Tools for AI Agents: Patterns That Make Agents Reliable
Most agent failures are tool-design failures. Practical patterns for naming, scoping, validating and securing tools so LLM agents choose and use them correctly.
-
19 Oct 2026 · 6 min
The Production RAG Checklist: 25 Things to Get Right Before Launch
A practical checklist for building retrieval-augmented generation systems that work in production — ingestion, chunking, retrieval, grounding, permissions, evaluation and operations.
FAQ
Frequently asked questions
What does Dhii do?
Dhii is an independent AI engineering studio. We design, build and run agentic automations, AI agents, RAG knowledge assistants, chatbots and voice agents, and we fine-tune, train and self-host models — for teams worldwide.
Do you work with international clients?
Yes. We are based in Mumbai (IST, UTC+5:30) and work remotely with clients in other time zones, with overlap hours agreed at the start of each project. Communication is in English.
How does an engagement start?
With a short call and then usually a one- to two-week Discovery Sprint: we map the workflow, prototype on your real data and give you a fixed, written quote for the build.
Which AI models and tools do you use?
Whatever performs best on your evaluation set at a sensible cost — OpenAI, Anthropic and Google models, or open-weight models such as Llama, Qwen, Mistral and Gemma served with vLLM or Ollama. We are vendor-neutral.
Can the AI run on our own infrastructure?
Yes. We deploy private, self-hosted language models on your servers or cloud account so data never leaves your environment. We run this stack ourselves at chat.dhii.ai.
Who owns the code and models?
You do. All code, prompts, datasets, trained weights and documentation are delivered to you, subject only to the licence terms of any third-party base models.
Start a project
Have a workflow that should run itself?
Tell us what you want automated or built. You get a written assessment of feasibility, risks and a fixed quote — usually within two business days.