One process at a time, measured at every step
We start small, on one process and your real data, and only build what the numbers support. Here is exactly what happens from the first call to the monthly report, and what we need from you at each stage.
Book a discovery callThe four stages
- 1
Discovery call
We look at one process with the person who runs it: the steps, the systems it touches, how many cases a month, and what success would mean in numbers. If AI isn’t the right fit, we say so on the call.
You get
- A short note on whether AI can take the process over
- The approach we would take and what it connects to
- An estimate for setup and monthly running costs
We need from you
- A call with someone who does the work
- Rough monthly volumes
- 2
Proof of concept
We build a working version on a sample of your real, past cases, including the messy ones, and measure it against pass marks we agree with you first. You judge results on your own data, not slides.
You get
- A working prototype you can try
- A results report on your real cases
- A fixed quote and plan for the full build
We need from you
- A sample of real past cases
- Read-only access to one system
- Someone to review results with us
- 3
Build and launch
We connect the system to your tools, build the approval screens your team needs, and test every change against your cases. Launch happens in stages: drafts for your team to approve first, then a share of live work, then more as the numbers hold.
You get
- The system running in production
- A test set of your cases, run on every change
- Documentation and training for your team
We need from you
- Access to the systems it connects to
- Someone to sign off at each demo
- Your IT or security review
- 4
Run and improve
We watch the system every day, review wrong answers, update it when your policies or products change, and keep usage costs in check. Once a month you get a plain report on what the AI did and what it cost.
You get
- Monitoring and fixes
- Model and content updates
- A monthly report and review call
We need from you
- A short monthly review call
- A heads-up when processes change
ExampleMonthly report, August 2026
Customer support assistant
- Conversations handled
- 4,812
- Resolved without a person
- 71%
- Handed to your team
- 1,396
- Average first reply
- 9 sec
Resolved without a person, by channel
23 wrong answers found in reviews. All fixed and added to the test set.
Usage cost ₹38,400, 6% under budget.
Next month: add the new returns policy and Tamil replies.
Our tech stack
Tools and technology we build with
We pick the right tool for each job, not the newest one. These are the models, platforms and systems we use most.
- AI models
- Claude
- OpenAI
- Gemini
- Llama
- Mistral
- Sarvam
- AI frameworks
- LangChain
- LangGraph
- LlamaIndex
- Hugging Face
- Voice
- Exotel
- Twilio
- Plivo
- Deepgram
- ElevenLabs
- Messaging
- WhatsApp API
- Gupshup
- Interakt
- Gmail
- Outlook
- Business apps
- Tally
- Zoho
- Salesforce
- HubSpot
- Freshdesk
- Zendesk
- Shopify
- Front-end & mobile
- React
- Next.js
- TypeScript
- Tailwind
- React Native
- Flutter
- Back-end
- Python
- FastAPI
- Node.js
- Go
- Data
- Postgres
- pgvector
- Redis
- MongoDB
- Qdrant
- Pinecone
- Cloud & DevOps
- AWS
- Azure
- GCP
- Docker
- Kubernetes
- Terraform
- GH Actions
- Testing
- Langfuse
- Playwright
- Pytest
- Sentry
- Grafana
How we run every project
- Demos, not status reports
- You see the system working on your real data throughout, and can change direction early.
- Fixed price for each stage
- Each stage is quoted in writing before it starts. If we underestimate, that’s on us.
- You own what we build
- Code, prompts, test sets and data belong to you. Everything is documented and handed over.
- One senior lead throughout
- The person on your discovery call stays on your project through launch and after.
- No lock-in
- Systems run in your cloud account or ours, on models you can switch. You can take it in-house at any time.
- An honest no
- Some processes aren’t ready for AI. We would rather tell you than build something that won’t be used.
Who you’ll work with
Project lead
Your single point of contact. Runs discovery, owns the plan and the weekly demo.
AI engineer
Builds the agent, its prompts and its checks, and chooses the right models.
Integration engineer
Connects the system to your CRM, ERP, telephony or WhatsApp.
Quality reviewer
Builds the test set from your cases and reviews every failure by hand.
Support engineer
Watches the live system after launch and handles fixes and updates.
Questions about the process
Can we skip the proof of concept?
If the process is simple and well understood, sometimes yes. But for most projects the proof of concept is the cheapest way to find out whether the build is worth it, and it gives you a fixed quote based on real results.
What if the proof of concept doesn’t hit the pass marks?
You get the results report either way. Often it shows which part of the process AI can handle and which it can’t, and we suggest a smaller scope. If it isn’t worth building, we tell you, and you’ve only spent the cost of the proof of concept.
Do we have to sign a long contract for the monthly plan?
No. There’s no long lock-in, and you can take the system in-house at any time with full documentation.
How much of our team’s time does a project need?
Very little: some time from the person who runs the process to review results, plus access and sign-off from IT. We do the rest.