We tell you where AI helps, and where it doesn't.
Drish is an AI consultancy for companies that are past the demo stage and need something that holds up on a Tuesday afternoon in production.
Drish — from dṛś, to see
Most AI projects die between the demo and the org chart.
The demo was never the hard part. A model that impresses a steering committee still has to survive messy data, edge cases, and the person who has done the job manually for eleven years.
The bottleneck is rarely the model. It is usually an undocumented process, a system with no API, or an unclear answer to who owns the output when it is wrong.
We work in your systems, not around them. Every engagement ends with something running, documented, and owned by someone on your payroll.
What we do
Where AI actually helps
A short diagnostic across your workflows, data, and tooling. You get a ranked list of what is worth automating, what is not, and what would need to change first — with cost and effort attached to each.
From pilot to production
You have something that works in a notebook and stalls everywhere else. We handle evaluation, failure modes, human review steps, monitoring, and the integration work that makes it a real system instead of a demo.
Custom AI systems
Document search over your own material, assistants wired into internal tools, agents that handle a defined slice of a workflow. Built on your data, in your stack, with the accuracy measured rather than assumed.
Teams and guardrails
Training for the people who will use these tools daily, plus a usage policy, review process, and vendor position that legal and security can sign off on without a six-month detour.
Fit check
Describe a workflow that is costing you time. This page will run a language model on your own graphics card to match it against patterns we have seen, and tell you whether AI is likely to help. Nothing you type is sent anywhere.
Results will appear here.
How this works, since you will ask: an embedding model (mxbai-embed-xsmall, 25 MB) is downloaded to your browser and run through the WebGPU API on your own hardware. Your text is compared against sixteen patterns we wrote by hand. It is a similarity search, not a model reasoning about your business — the same technique underneath most "chat with your documents" products, shown without the curtain. If your browser has no WebGPU it falls back to WebAssembly and simply runs slower.
How an engagement runs
Look
Two to three weeksWe sit with the people doing the work, read the data, and map what actually happens. It ends in a written recommendation you can act on even if you never hire us again.
Build
Six to ten weeksOne system, chosen because it pays for itself. Shipped to real users in stages, with accuracy measured against a benchmark we agree on before we start.
Hand over
Two weeksDocumentation, runbooks, and working sessions with your team until they can change the system without calling us. Then we leave, and stay reachable.
Whether we are a fit
We tend to do good work when
- You have a specific workflow that is expensive, repetitive, and well understood internally.
- Someone senior is accountable for the outcome and can make decisions in the room.
- Your data is imperfect but it exists somewhere we can reach.
- You want your team to own the system afterward.
We are the wrong firm when
- You need something impressive to show the board next month.
- The goal is a headcount reduction plan rather than a working system.
- No one internally has time to be involved for a few hours a week.
- You want a vendor who stays embedded indefinitely.
Start with a conversation, not a proposal.
Tell us the workflow that is bothering you. If we think AI is the wrong tool for it, we will say so on the call — that answer is free and it saves everyone a quarter.
Email us- hello@mydrishinc.com
- Phone
- [your number]
- Based in
- [your city] — working with teams across North America
- Typical engagement
- [your range, e.g. $15k–$80k]