AI Division
AI that takes real work off your team.
We build agents, models and assistants that work inside the systems you already use, on tasks like order entry, invoice matching, ticket triage, approvals and reports. One strategist and two engineers take each use case from assessment to production, then stay to run it.
In short
What does the AI Division build?
AI agents, machine learning models, document reading pipelines and private knowledge assistants that work inside the systems you already use, such as Tally, your ERP, your CRM or WhatsApp. One strategist and two engineers take each use case from assessment to production, then stay to run it.
- How long does an AI project take?
- A 1 to 2 week assessment with a fixed scope, a 2 to 3 week pilot, then rollout to production and a monthly managed operations plan.
- What does it cost?
- USD 2 or INR 199 an hour under 2good.dev, with a 150 hour minimum. The pilot is paid only after you accept it against the measures agreed in the assessment.
- Which models do you use?
- Anthropic Claude, OpenAI, Google Gemini and open weight models such as Llama, Qwen and Mistral, chosen by task, data sensitivity and cost.
- Is our data used to train models?
- No. Work runs under enterprise agreements that rule out training on your data, or on open weight models on your own servers.
Services
What the AI Division delivers
Agents, machine learning, document reading and private assistants, built and run by small forward deployed pods.
- AI Agents and AutomationAgents that process orders, invoices, support tickets, approvals and reports.
- Machine Learning ModelsForecasting, scoring, classification and computer vision on your own data.
- Document IntelligenceInvoices, POs, KYC and contracts read and posted into Tally, your ERP or CRM.
- Private Knowledge AssistantsAnswers from your own documents and data, kept inside your company.
- Voice and WhatsApp AgentsCalls, bookings, reminders and order status in Hindi, Gujarati and English.
- AI Inside Your ProductsCopilots, smart search and MCP connectors in your apps, ERP or CRM.
- AI Governance and SecurityAI policies, EU AI Act and ISO 42001 readiness, and red team testing.
- Managed AI OperationsMonitoring, quality checks, cost control and model updates, under an SLA.
See it work
An agent at work, sketched.
Work arrives by email, WhatsApp and documents. The agent reads it, checks it against your rules and records, acts in your systems and sends anything uncertain to a person. Every step is logged.
How we deliver AI
Four steps. One accountable pod.
Every AI engagement follows the same path, with a clear gate between steps.
Step 1: Assess
ExampleUse case shortlist
Ranked by value, effort and data readiness
- Invoice entry into TallyHigh
- Support ticket triageHigh
- Weekly sales reportMedium
- Contract renewal trackingMedium
- Vendor onboardingLater
An illustrative view of a typical engagement, made without client data.
The delivery pod
One Forward Deployed Strategist. Two Forward Deployed Engineers.
The strategist owns the client, the process and the results. The engineers build, connect and deploy. QA is shared across pods. The pod works on Anthropic Claude, OpenAI, Google Gemini and open weight models, on Azure AI Foundry, AWS Bedrock, Google Vertex AI or your own servers.
Principles
How this division works
- Pick the work first
- We choose use cases by the hours they save and the errors they remove. The model is chosen after that.
- Your data stays with you
- When the data is sensitive, we deploy on your cloud, a private tenant or your own servers.
- Prove it before scaling
- Each pilot has success measures agreed up front, and it is judged against them before anything scales.
- We keep it running
- After launch, monitoring, quality checks and cost control continue on a monthly plan.
Toolkit
The stack we actually work in
- Models
- Anthropic ClaudeOpenAIGoogle GeminiLlamaQwenMistral
- Agents and retrieval
- LangGraphClaude Agent SDKOpenAI Agents SDKMCPLlamaIndexpgvectorQdrant
- Platforms
- Azure AI FoundryAWS BedrockGoogle Vertex AIYour own servers
- Data
- PostgreSQLDatabricksMicrosoft FabricdbtAirflow
Client work
Recent projects from the AI Division
- HealthTech · FDEAI medical imaging that reads X-rays, CT and MRI scans for a healthcare provider70%Faster diagnosis, with patient wait times down from hours to minutesRead the case study
- FMCG and Consumer Goods · O2EAI agent that reads supplier invoices and posts them into Tally for a distributorA regional FMCG distributor in Gujarat supplying kirana stores and modern tradeRead the case study
- Oil and Gas · FDEDigital permit to work and HSE incident reporting for an oil and gas terminalA coastal petroleum products storage and dispatch terminal on the west coast of India, handling tank farm, gantry loading and jetty operationsRead the case study
Questions
Good to know
Where should we start with AI?
With a 1 to 2 week assessment. We map your processes, rank use cases by value and readiness, and agree success measures before anything is built.
Do we pay for a pilot that does not work?
The pilot is paid only after you accept it against the measures agreed in the assessment.
Can AI run on our own infrastructure?
Yes. We deploy on Azure AI Foundry, AWS Bedrock, Google Vertex AI, private tenants or your own servers with open weight models.
Who looks after the system after launch?
Our managed AI operations plan covers monitoring, quality checks, cost control and model upgrades, with an SLA.
Start a conversation
Tell us what you are trying to build.
Share the problem in a short call. You leave with a clear view of the approach, the team and the cost.
A specialist replies within one working day. The first conversation is free.