AI Agent Development & Workflow Automation India

FILE 01 / WHAT IS AN AI AGENT

A worker, not a chat window.

DEFINITION

An AI agent is a language model plus a set of tools it's allowed to use, plus a goal. You give it something like "close out this compliance filing" or "reconcile these two spreadsheets" and it plans the steps, picks the tools (an API, a database, a search, a file writer), executes them one by one, checks its own output, and only comes back to you when the job's done or when it needs a decision it can't make.

Chatbots have one loop: user talks, bot replies. Agents have many loops: think, act, observe, think again. Which is why they're slower and more expensive per query. But capable of finishing work a chatbot can't touch.

RUN LOG How an agent runs a task

✓ Goal received — "Find all PII in our AWS S3 bucket and mask it."
✓ Plan generated — List files → scan each → classify PII → mask → log → report.
✓ Tools called in a loop — S3 API → OCR → PII classifier → mask → write log.
✓ Human check on high-risk items — Ambiguous cases queued for DPO approval.
✓ Report + audit trail — Every step logged. You can replay any decision.

FILE 02 / PROOF OF WORK

Dcomply: 51 agents running compliance

DPDP Act 2023 changed everything. Dcomply is our answer. Compliance agents that read regulations, scan data, and file returns without a human in every loop.

01 THE PROBLEM

DPDP penalties up to ₹250 crore. Data scattered across 50+ systems.

Where is your PII? S3, GDrive, Postgres, Slack messages, WhatsApp exports, old CSVs on someone's laptop. Manual audits take weeks and miss things.

02 WHAT WE BUILT

51 auto-connectors + India-tuned PII discovery.

Each connector is an agent. Scans its source, classifies data, flags PII (Aadhaar, PAN, GSTIN, UPI, bank accounts), and reports. Another agent handles DSR (Data Subject Request) fulfilment end-to-end.

03 THE RESULT

Discovery from weeks to hours. 80+ modules in one platform.

DPDP consent, DSR, PII discovery, breach notification, GSTR filings, ISO 27001 audit trail, sector-specific rules. The agents do the scanning, filing, and reporting. DPO signs off on the edge cases.

STACK-01 Agent Framework

LangGraph + Claude

Stateful multi-step flows

STACK-02 Tool Calling

51 API connectors

S3, GDrive, Slack, Postgres…

STACK-03 PII Classifier

Custom NER model

Aadhaar, PAN, GSTIN, UPI

STACK-04 Guardrails

Human-in-loop + audit

Every step logged, replayable

See Dcomply live →

FILE 03 / SIX AGENT PATTERNS WE SHIP

These are the ones that actually work in production.

We won't sell you an agent for something an agent isn't ready to do yet.

AGT-01

Compliance & Audit Agents

Read a regulation, cross-check your data, flag gaps, draft the filing, log the trail. Works for DPDP, GST, RBI KYC, SEBI, ISO 27001, sector rules.

LIVE IN: DCOMPLY · 80+ MODULES

AGT-02

Document Processing Agents

Read invoices, contracts, IDs, medical reports. Extract structured data. Cross-check against your DB. Route by outcome. Handles handwritten, scanned, and multi-language.

USES GPT-4O VISION + CUSTOM OCR POST-PROCESSING

AGT-03

Ticket Triage & Routing

Read the ticket. Search past resolutions. Try to answer. If it can't, categorise + tag + route to the right queue with context. Cuts first-response time 60-80%.

SITS INSIDE FRESHDESK, ZENDESK, INTERCOM

AGT-04

Sales & Outbound Agents

Research the prospect. Pull context from CRM + LinkedIn + past emails. Draft a personalised outreach. Schedule follow-ups. Escalates to human on positive replies.

FEEDS INTO YOUR CRM. NEVER FULLY AUTONOMOUS

AGT-05

Data Reconciliation Agents

Match rows between two systems that don't quite agree. Bank statements vs. ledger. GSTR-2A vs. purchase register. Vendor list vs. payment records. Flag mismatches with reasoning.

HANDLES FUZZY NAMES, DATES, AMOUNTS

AGT-06

Research & Reporting Agents

Pull data from 5-10 sources, cross-check, summarise into a report. Monthly compliance summaries, competitor tracking, market briefs. Same shape as Dcomply's monthly reports.

DELIVERED AS PDF, SLACK, OR DASHBOARD

FILE 04 / THE PART MOST AGENCIES SKIP

How we stop agents from doing damage.

An agent that can call APIs can also delete rows, send bad emails, or file a wrong return. The four guardrails below are the ones we ship on day one. Because we're the team that runs it after launch, and we don't want the on-call ping.

01

Whitelisted Tool Access

Agent can only call tools you explicitly allow. Everything else returns "not permitted".

02

Dry-Run for Destructive Actions

Any delete, send, or file action shows a preview. Human confirms before it fires.

03

Hard Spending Caps

Model API cost cap. Tool call cap. Outbound message cap. Agent stops when it hits any of them.

04

Full Replayable Audit Log

Every thought, tool call, input, output, decision. You can replay any step to see why it did what it did.

FILE 05 / PROCESS & PRICING

First agent live in 8-12 weeks.

Priced by phase. No surprises.

WEEK 1

Workflow Audit

We map the task, the tools, the risk points, and the human-approval gates.

FREE

WEEKS 2-4

Prototype Agent

One agent, one workflow, dry-run only. You watch it work on real data.

₹2-4 L

WEEKS 5-12

Production Build

Full integrations, guardrails, audit logging, human-in-loop, monitoring, staging + prod.

₹8-25 L

ONGOING

Operate

Model API costs, monitoring, tool updates, quarterly evals, incident response.

FROM ₹45K/MO

REF-A Our Agent Stack · No Lock-In

LangGraph · CrewAI · AutoGen · Custom Python orchestration · OpenAI Assistants API · OpenAI GPT-4o · Anthropic Claude Opus 4 · Llama 3.3 (self-hosted) · Mistral Large (on-prem) · n8n · Zapier · Custom webhooks · REST / GraphQL adapters · Kafka · LangSmith · Langfuse · OpenTelemetry · PagerDuty · CloudWatch + Grafana

We recommend the stack in the prototype phase and lock it in with you before production build. If your team already runs Kubernetes, Postgres, or a specific cloud. We work inside it. No forced migrations.

REF-B Agents by Industry

REF-C Related AI Capabilities

FILE 07 / FIELD MANUAL — FAQ

What buyers ask us about agents

What is an AI agent?

An AI agent is a system that uses a language model to plan a sequence of steps, use tools like APIs or databases, and finish a task on its own. A chatbot answers questions. An agent does work. Read a 500-page regulation and pull out the applicable clauses. Scan 51 systems for personal data. Draft an email, check the calendar, book the meeting. The agent decides which tool to call next until the job is done.

How is an AI agent different from RPA?

RPA follows a script that a human wrote. It breaks the moment a screen layout changes or an input format shifts. An AI agent uses a language model to figure out what to do next, so it handles variation. Trade-off: RPA is faster and cheaper for rigid workflows. Agents are the right choice when the work involves reading unstructured data, making judgment calls, or working across many systems.

How much does an AI agent cost to build?

Prototype phase (one agent, one workflow) is ₹2-4 lakh. Full production build with multiple tools, guardrails, human-in-the-loop escalation, monitoring, and integrations is ₹8-25 lakh. Ongoing operations start at ₹45,000/month and include model API costs, monitoring, tool updates, and quarterly evals. Feasibility call is free.

How long does it take to build an AI agent?

A prototype agent runs on your data in 3-4 weeks. Production build takes 5-8 weeks after that. Total: 8-12 weeks. Complex multi-agent systems with 10+ tools take 12-16 weeks.

Which agent frameworks do you use?

LangGraph for stateful multi-step workflows. LangChain agents for simpler patterns. AutoGen for multi-agent conversations. OpenAI Assistants API and Anthropic Claude tool-use for shipped features. We pick based on what the workflow needs. No religious preference.

How do you stop an AI agent from doing damage?

Four layers. First, permissions. The agent can only call tools you explicitly whitelist. Second, dry-run mode. Every destructive action is previewed and needs a human confirmation. Third, spending caps. Hard limits on model API costs, database writes, and outbound messages. Fourth, full audit logs of every step the agent took, why, and what it changed.

Can the agent work with our existing systems?

Yes. Dcomply already has connectors to S3, GDrive, Slack, Postgres, MySQL, Zoho, HubSpot, Freshdesk, WhatsApp, Gmail, Notion, and 40+ more. If your system has an API, we can wire an agent to it. If it doesn't, we can build a browser-automation layer as a last resort.

What tasks are AI agents good at right now?

Reading long documents and extracting structured facts. Cross-checking data across systems. Ticket triage and routing. Email drafting with context from your CRM. Compliance filings that follow rules. Research tasks that need multiple sources. Reporting that pulls from several places. What they're still bad at: tasks with high physical stakes or heavy real-time judgment. We won't sell you an agent for those.

Have a workflow you'd like to hand off?

Ask Deci (bottom-right) about the work you'd like an agent to do, or drop us a line via the contact form. We'll tell you if it's a fit, what it costs, and what your team still needs to own.

Chat with Deci See other AI capabilities →