AKSHAT_ Let's talk ↗

AI AUTOMATION / N8N SPECIALIST

I build AI systems that do the work.

I design autonomous workflows, AI agents and API-driven automations that turn repetitive processes into systems that run without constant human intervention.

Featured Work Approach
BASED IN INDIA AI × AUTOMATION × SYSTEMS 2026

01 / SELECTED WORK

Systems, not demos.

A project is more useful when you can see the problem, the architecture and what changed after automation.

n8n workflow: a schedule trigger fetches news and Google Trends data, merges and reduces it, an AI agent picks the strongest topic and drafts a tweet, and a final step posts it to X
01
AI AUTOMATION
OPEN CASE STUDY See how the system works ↗

ZeroTouch ✓ n8n Library Accepted

Autonomous X content pipeline that discovers live trends, selects a topic, drafts a post and publishes it on a daily schedule.

n8nAI AgentGroqX API
n8n workflow for RELAY: Telegram intake, Brain and Critic agents, memory, and Slack escalation
02
AI SUPPORT AGENT
OPEN CASE STUDY See how the system works ↗

RELAY

A Telegram support agent that answers from a real knowledge base, verifies its own replies, and opens a tracked Slack ticket when it can't help.

n8nGroqPineconeSupabaseTelegramSlack
Workflow showing a document ingestion pipeline into a vector store and an AI agent that answers questions from those documents
03
DOCUMENT AI
OPEN CASE STUDY See how the system works ↗

DocuMind

An assistant that reads a company's documents once and answers questions about them instantly, with a built-in watchdog that flags failures before anyone notices.

n8nPineconeGemini EmbeddingsGroqSlack

02 / APPROACH

From first call to running system.

Every engagement follows the same shape, whether it's a single workflow or a full agent pipeline.

01

Discovery call

A short call to understand what currently happens manually and where the bottleneck actually is.

02

Scope & proposal

A written scope of the workflow, the tools involved, and a timeline — before any build work starts.

03

Build & test

The automation is built in stages, with real data, and checked against edge cases and failure modes.

04

Handover & support

Documentation, access to the workflow, and a support window to fix anything that comes up post-launch.

03 / CONTACT

Have a repetitive process worth automating?

Tell me what currently happens manually. I'll help break it into a workflow that can be automated.

EMAIL akshat4work2004@gmail.com ↗

Currently taking on new automation projects

CASE STUDY / 01

ZeroTouch

A fully autonomous n8n workflow that posts a trend-aware tweet every day with zero manual input.

✓ n8n Library Accepted
n8nNewsAPI / GNewsGoogle TrendsGroqX APIOAuth 1.0a
DAILYscheduled automation

1) PROBLEM

Publishing trend content was manual and inconsistent.

Finding a strong topic every day required checking multiple sources, deciding what mattered, drafting copy and posting manually. The process was repetitive and easy to skip on busy days, which hurt consistency.

2) SOLUTION

Built a fully automated trend-to-post pipeline in n8n.

The system fetches NewsAPI/GNews and Google Trends in parallel, normalizes both feeds, lets a Groq-powered AI agent pick the strongest topic with rolling-week memory, drafts the tweet, and publishes to X via OAuth 1.0a.

Production hardening included fixing X permission scopes and resolving API credit/billing edge cases so the workflow can run reliably every day.

3) IMPACT

Daily publishing with zero manual steps.

  • Working hours saved~32 hrs/month
  • Estimated cost saved~$720/month
  • Publishing consistencyfrom ad hoc to daily

01 / WORKFLOW

Full workflow diagram for the trending topic bot

02 / CORE CAPABILITIES

Data sources News + trends
AI agent Analyzes trends
Tweet draft AI writes copy
Auto-post Published to X

Runs on a schedule — fully automated, no manual steps

REUSABILITY

Same architecture, multiple channels.

The same retrieve → normalize → evaluate → draft → publish pattern can be adapted for LinkedIn, Instagram, curated newsletters, competitive monitoring and daily digests without redesigning the system from scratch.

BUILT WITH n8n + AI Let's talk ↗

CASE STUDY / 03

DocuMind
Ask your documents.

A retrieval-augmented assistant that reads a company's documents once and answers questions about them in plain language, the way you'd ask a colleague.

n8nGoogle DrivePineconeGemini EmbeddingsGroqSlack
SELF-HEALINGfailures alert Slack automatically

1) PROBLEM

Teams were losing time searching scattered documents.

People had to manually scan PDFs and folders to answer internal and customer questions. Responses were slow, duplicated across team members and occasionally inconsistent when document versions changed.

2) SOLUTION

Implemented a RAG assistant with automatic document sync.

DocuMind ingests updated files from Google Drive, chunks content, creates embeddings with Gemini, stores vectors in Pinecone, and uses a Groq-powered AI agent to answer questions grounded in retrieved context.

A dedicated error workflow sends plain-language Slack alerts on failures, and re-upload logic removes stale vectors to prevent duplicate or outdated answers.

3) IMPACT

Faster answers with fewer interruptions.

  • Working hours saved~54 hrs/month
  • Estimated cost saved~$1,350/month
  • Average response timefrom ~25 min to <3 min

01 / THE AUTOMATION

n8n workflow: document ingestion into a Pinecone vector store on one side, and a webhook-triggered AI agent that searches the database and answers the user on the other

RELIABILITY LAYER

Monitored so failures are visible immediately.

When ingestion, retrieval or model calls fail, Slack alerts identify the exact broken step and file so issues can be fixed quickly before trust in the assistant drops.

BUILT WITH n8n + RAG Let's talk ↗

CASE STUDY / 02

RELAY

A self-verifying AI support agent for Telegram: it answers from a real knowledge base, checks that answer against evidence, and escalates to a human with a tracked Slack ticket when it can't help.

n8nGroqPineconeGemini EmbeddingsSupabaseTelegramSlack
BRAIN + CRITICtwo-model verification before send

1) PROBLEM

Support bots either hallucinate or fail silently.

A single LLM call invents policies and order statuses with confidence. Duplicate webhooks double-reply, bursts drain API budget, and a real escalation gets duplicated because the bot doesn't know it already raised a ticket.

2) SOLUTION

Built a Brain + Critic pipeline with human handoff in Slack.

The conversational workflow receives a Telegram message, dedups and rate-limits it, retrieves memory, drafts a grounded answer with four tools only, then lets an independent Critic approve, rewrite, or escalate.

If it can't help, RELAY opens a tracked Slack ticket and a teammate closes it with a Slack slash command — no separate admin panel.

3) IMPACT

Verified answers, clean escalations.

  • Resolved without a human89%
  • Incorrect order-status replies0 across the test set
  • Average response time~3.1s end to end

01 / WORKFLOW

Full RELAY n8n workflow overview Trigger, dedup and rate limiting nodes Workflow Brain agent with knowledge base, order lookup and ticket tools Independent Critic model verifying the Brain draft Escalation path creating a Slack ticket

Full system overview

01 / 05

02 / CORE CAPABILITIES

Telegram Dedup + rate limit
Brain Grounded tools only
Critic Evidence check
Reply / Slack Send or escalate

Every reply is verified before it ships — unresolved issues become tracked Slack tickets

RELIABILITY LAYER

Built for noisy webhooks, not just the happy path.

Webhook dedup, fail-open rate limiting, ticket-already-open checks, and fallback Telegram replies mean a crash or API blip degrades to an apology — never silence, never a duplicate ticket, never a second “we're escalating.”

BUILT WITH n8n + AI Let's talk ↗