0
PIPELINE: JAYANT.CHAURASIA · STATUS: LIVE · LISBON

I build agents that do the boring work.

Jayant Chaurasia — Automation Engineer. AI agents, n8n pipelines & internal tools for business teams. If it happens twice, it becomes a workflow.

SCROLL TO EXECUTE · ⌘K FOR COMMANDS
node:02 parse.story QUEUED

From doing the work
to deleting it.

I didn’t study computer science. I studied management (CEMS), then ran fleet operations — and got so annoyed by repetitive work that I automated my own job. Six months later, the company made that my actual job.

2023 – 2025
2023 – 2025masters.cems() — Management, Strategy major. NOVA Lisbon + NHH Norway.
04 / 2025
04 / 2025ops.trainee() — 4 depots, 250+ vehicles, 10+ vendors. Found the bottlenecks the hard way: by living them.
10 / 2025
10 / 2025promote("Automation Engineer") — turned my own backlog of annoyances into production automations.
today
todayagents.ship() — 48 workflows built, AI knowledge platform live for 700 people, customer-support agents in build.
node:03 metrics.aggregate QUEUED

Output, measured.

Numbers from the company n8n cloud instance and production systems — not vibes.

0
n8n workflows built
0
live in production
0
execution success rate
0
attributed savings
0
FTE of manual work removed
0
people served by my AI bot
0
rows extracted from 6 reverse-engineered APIs
0
request → production cycle
node:04 runs.map(projects) QUEUED

Recent executions.

Each one started as somebody’s repetitive Tuesday.

AI Knowledge Platform

● PROD

RAG Slack bot answering operations & support questions for a 700-person workspace. Ingestion pipeline, feedback capture, health monitoring, daily KPI reports — hallucination-tested with adversarial multi-agent audits before rollout.

staff questions answered in seconds, not escalations
n8nSupabase pgvectorGeminiSlackRAG

Customer-Support AI Agents

◐ IN BUILD

Extending the knowledge platform from answering staff questions to handling support conversations: retrieval grounding, intent routing, evaluation loops, escalation-to-human design.

digital workers for the support queue
agentsRAGeval loopsescalation design

OEM Warranty-Claim Agent

● PROD

End-to-end agent that reads repair invoices (Gemini extraction + fault classification) and files OEM warranty claims in a dealer portal via Playwright — first real claim filed June 2026, replacing a fully manual multi-system process.

money recovered while humans sleep
PythonPlaywrightGeminin8n micro-service

Invoice Intelligence Pipelines

● PROD

8-workflow document-extraction family (Gemini vision OCR) across 3 operating regions, plus finance automation fanning out over 29 legal entities with composite-key + hash deduplication.

invoices that process themselves
at 1,500 invoices/mo → 75 hours of manual review removed, every month
n8nGemini OCRSpendeskdedup logic

API Reverse-Engineering Portfolio

✓ DONE

6 undocumented enterprise portals reverse-engineered in pure Python: auth flows, token lifecycles, WAF backoff, session-scoped endpoints — 60,000+ catalogue rows delivered where browser automation was the wrong tool.

data nobody believed could be exported
Pythonrequeststoken authWAF handling

Job-Search Agent System

● PERSONAL

Multi-phase personal agent pipeline — discovery → research → fit-scoring → documents → dashboard — pairing LLM agents with deterministic Python validators behind a 180-test suite. Agents propose, code verifies.

my own repetitive work, removed
Claude agentsSQLitepytestorchestration

Consumer Mobile App

✓ QA-PASSED

Campervan-rental app built end-to-end in React Native + Expo (29 screens): hand-built API client against a reverse-engineered auth flow, in-app Gemini assistant with intent routing, Datadog RUM, release builds on real devices.

proof the “automation guy” ships software too
React NativeTypeScriptExpoDatadog
node:05 sandbox.interactive QUEUED

Don’t read about it.
Wire it yourself.

This is a working miniature of my actual job. Drag the nodes. Connect the loose one (click an output port ●, then an input port). Then hit RUN and watch the data flow — failures, retries and all.

› sandbox ready — wire the dashed node for full coverage, or just hit RUN.
node:06 stack.load QUEUED

The toolkit.

AI-first by default — my development environment is an agent, not an editor.

n8nClaude CodeGeminiPythonTypeScriptSupabasePlaywrightRAG n8nClaude CodeGeminiPythonTypeScriptSupabasePlaywrightRAG
MCP serversSlack APIsSQLAirtableMetabaseAWS AthenaRESTwebhooksOAuth MCP serversSlack APIsSQLAirtableMetabaseAWS AthenaRESTwebhooksOAuth
node:07 human.handoff QUEUED

Escalate
to human.

This is the only step I didn’t automate. On purpose.

chaurasiajayant03@gmail.com
linkedin --connect send --email