Resources
Reading for the people building and running AI.
Practical guides on choosing, building and governing agents, and longer articles on the questions that take real thought. Written by the AksharAI team, dated, and kept up to date.
Start here
- Guide · Getting started · 6 minChoosing your first AI agent: a one page testMany first AI projects stall because the wrong process was picked. A one page test helps you choose a task that is frequent, rule based and easy to check, with a person approving the result.Read it
- Guide · Building agents · 6 minAI agents, chatbots and RPA: what is the difference?A chatbot answers questions, RPA repeats fixed steps in software, and an AI agent works through a task using judgement within rules. Most businesses need more than one, each for the work it suits.Read it
- Guide · Governance · 5 minWhere a person must approve: designing human checkpointsHuman review only works when it is placed at the right step and the reviewer has what they need to decide. That means finding those steps, designing the review screen and keeping reviewers alert.Read it
28 of 28 pieces
- Guide · Use casesAI agents for sales teams: CRM updates, visit briefs and follow upsSales agents handle the work around selling: logging visits and calls in the CRM, preparing a brief before each visit and drafting follow ups. The rep keeps the conversation, the price and the commitment.6 min read
- Guide · Building agentsAI agents, chatbots and RPA: what is the difference?A chatbot answers questions, RPA repeats fixed steps in software, and an AI agent works through a task using judgement within rules. Most businesses need more than one, each for the work it suits.6 min read
- Guide · Use casesAI ticket triage for support teamsAI triage reads each new ticket, works out what it is about and how urgent it is, adds missing details and puts it in the right queue. Support leads set the rules and review the cases the agent is unsure about.6 min read
- Guide · Use casesKYC document extraction for NBFCs and banksAn extraction agent can read KYC documents, pull out the fields, cross check them and flag mismatches. The KYC officer still verifies identity and decides, under the RBI's KYC Master Direction and the lender's own policy.6 min read
- Guide · Data and integrationsRetrieval augmented generation (RAG) for company documents, in plain EnglishRAG lets an AI model answer from your own documents. The system finds the most relevant passages first, gives only those to the model, and asks for an answer that cites them, so people can check the source.6 min read
- Guide · Use casesWhat an AI IT helpdesk agent can do for your IT teamAn IT helpdesk agent takes requests in Teams, WhatsApp or email, collects the details IT needs, walks staff through common fixes and raises filled in tickets for the rest. Access and security stay with IT.6 min read
- Guide · GovernanceWriting a company AI usage policyAn AI usage policy tells staff which AI tools they may use, what data they may put into them, which outputs need checking and who to ask. A good one is short, plain and reviewed every few months.6 min read
- Article · GovernanceAgents and the DPDP Act: personal data in prompts, logs and outputsWhen an agent reads a borrower's message, looks up a patient record or drafts a reply to a customer, personal data moves through prompts, model calls, logs and outputs. Each of those flows carries duties under the Digital Personal Data Protection Act, 2023 and the DPDP Rules, 2025, and each has practical design choices that meet them.8 min read
- Guide · Data and integrationsAI for Tally users: what an agent can and cannot doMost Indian businesses keep their books in Tally, and most of the repetitive accounts work happens around it. An agent helps a Tally team most with reading documents, reconciling and answering questions from the books. It connects through Tally's own interfaces and leaves posting and payments to people.9 min read
- Guide · Building agentsChoosing a model provider for your agentsThere is no single best model. The right choice depends on the task, the languages your users write in, the volume, the speed you need and where your data is allowed to go. Test two or three candidates on your own cases, and design the agent so you can switch later.8 min read
- Guide · Getting startedChoosing your first AI agent: a one page testMany first AI projects stall because the wrong process was picked. A one page test helps you choose a task that is frequent, rule based and easy to check, with a person approving the result.6 min read
- Guide · Data and integrationsConnecting an agent to Tally, Zoho and your ERPAn agent is only useful if it can see your purchase orders, ledgers and customer records. Tally, Zoho and larger ERPs each connect in their own way, and in every case write access should stay narrow.5 min read
- Article · Use casesCopilots for finance teams: invoice matching, GST reconciliation and month endFinance teams in India spend a large share of each month matching documents, chasing suppliers and reconciling purchase registers with GST data. Three of those jobs suit a copilot, provided each is designed with limits that keep the books under a person's control.7 min read
- Guide · Data and integrationsDocument extraction accuracy: what to measure and howVendors quote accuracy figures for document extraction, but one number hides what matters. Measure extraction on your own documents, field by field, and design the process so the inevitable errors are caught.8 min read
- Guide · Building agentsEvaluating an AI agent before launchAn agent that looked good in a demo can still fail on the cases that matter. Test it on your own past work, score the results against agreed grades and set the launch bar before anyone relies on it.9 min read
- Guide · Data and integrationsGiving an agent your documents without leaking themAgents become useful when they can read your own documents. They become risky when they show those documents to the wrong person. Permissions, the choice of model provider, personal data handling and leak tests before launch decide which.5 min read
- Guide · Use casesHR policy assistants that employees trustAn HR policy assistant can take a large share of routine questions off the HR team, but only if employees trust its answers. That trust comes from a few design choices: the right policy per person, sources shown, privacy, and clear limits.8 min read
- Article · Building agentsLow code for AI: what it does well and where you still need engineersLow code builders make it possible for the people who understand a process to shape the AI that supports it. They do not remove the need for engineers. Some parts of an AI application suit a visual builder well, and others still need code.7 min read
- Guide · Getting startedMeasuring whether an AI agent is worth keepingEvery agent should earn its place. Set a baseline before launch, measure four things each week, count the full cost, and decide at ninety days whether to keep, change or switch off the agent.5 min read
- Guide · GovernancePrompt injection and safe tools for business agentsAny agent that reads text from outside, such as customer emails, supplier invoices or web pages, can be fed instructions hidden in that text. Narrow tools, approvals, rule checks and regular log review limit the damage.8 min read
- Guide · GovernanceRunning AI on premise or in a private cloudSome data should not leave your servers. Hosted, private cloud and on premise deployments differ in control, cost and model quality, and each needs its own hardware, skills and trade offs.5 min read
- Article · Teams and skillsThe forward deployed model: putting builders inside the businessMost AI projects are built at a distance from the work they change. The forward deployed model puts engineers and strategists inside the team for the length of the build. It rests on two roles and a set rhythm, fits some work poorly, and needs people trained for it.8 min read
- Guide · Getting startedWhat it costs to run an AI agentThe cost of an agent has four parts: the platform, model usage, the build, and the people who review its work. Model usage can be estimated from your own volumes, and review time is the cost teams most often leave out.8 min read
- Guide · Use casesWhatsApp agents for D2C brands: orders, returns and CODFor many Indian D2C brands, WhatsApp is the main support channel. An agent handles order status, returns and cash on delivery confirmation well. Refunds, damaged parcels and complaints need a person, and the WhatsApp Business Platform rules shape the whole design.9 min read
- Guide · Use casesWhatsApp support agents for Indian customers: Hindi, Gujarati and HinglishIndian customers write to brands in Hindi, Gujarati, English and every mix in between, often in Latin script. A useful agent handles those languages, looks up orders, follows WhatsApp's rules and hands over cleanly to your support team.5 min read
- Guide · GovernanceWhere a person must approve: designing human checkpointsHuman review only works when it is placed at the right step and the reviewer has what they need to decide. That means finding those steps, designing the review screen and keeping reviewers alert.5 min read
- Article · Getting startedWhy most AI pilots in Indian companies stall after the demoA good demo proves very little about whether an AI tool will survive real invoices, real customers and real month ends. Pilots usually get stuck in the same five places, and the work that gets them moving again is ordinary and practical.8 min read
- Guide · Building agentsWriting a prompt flow your team can maintainA prompt that one person wrote in a weekend becomes a problem the day that person goes on leave. Small steps, instructions in plain language and tests that run before each change keep the flow maintainable by the whole team.5 min read
Also useful
More ways to learn
Programme
Forward Deployed Professional programme
500 hours, two tracks, live project work with mentors.
Learn moreAnswers
Frequently asked questions
Models, data, pricing, deployment and how to start.
Learn moreCode Colonies
The wider library
Research and guides on software, consulting and compliance from the Code Colonies team.
Visit codecolonies.comSee it on your work
Bring one slow, repetitive process. We will show you an agent doing it.
A working session with someone who builds these every week.
Or get a ballpark proposal in two minutesWhat happens next
- Step 1A short call about one process and the systems involved
- Step 2A demo of an agent working on a version of your task
- Step 3A clear view of fit, approvals and what it would take