Shipping production GenAI — from EdTech to enterprise aviation.
DEPARTURES
FLIGHT · DESTINATION
STATUS
BV 2003
BOARDING
AI 0003
ON TIME
I build AI agents
that reason freely —
and rules that never guess.
Applied AI Engineer with 3 years building multi-agent orchestration, enterprise AI platforms and cloud-native systems. Right now at Torvalds, I'm automating airline operations where every legality verdict has to be explainable to a regulator.
The short version, for the 60-second read.
EASA, FAA Part 117, ICAO and DGCA rules encoded as versioned, testable rule sets.
End-to-end AI products built: dCortex, Hitwit, TalentFlow, Flo360 and Arivu.
LLM output never decides legality. Every verdict traces to an explicit, auditable rule.
How I work
Probabilistic models go where judgment helps. Deterministic code goes where safety matters. I'm careful about where the line between them sits.
What I'm best at
Multi-agent orchestration, long-context agent memory, MCP tool servers, RAG, and serverless backends on AWS.
dCortex — an AI operations platform for airlines.
THE PROBLEM Before a crew member flies, someone has to prove the flight is legal: duty limits, rest windows, carrier-specific overrides, and the rules of several regulators at once. Tools like Jeppesen and AIMS do this. dCortex does it too, built from scratch, with AI agents handling the operational work around it.
THE KEY DECISION
Keep the LLM away from the verdict. Agents read, plan and act, while a typed DSL rules engine makes the call,
so every legal / not legal
verdict can be audited.
DETERMINISTIC LANE
Rosters · NOTAMs · PDFsraw ops documents
VLM Parserschema-validated output
DSL Rules EngineEASA · FAA 117 · ICAO · DGCA
Legality verdicttraced to an explicit rule
typed data contracts
PROBABILISTIC LANE
Mem0⁺ memorypermission-scoped, namespaced
Agent orchestratormulti-layer, tool-calling
MCP serversportable capabilities
Airline APIslive data · work orders
Rules as data, not code
Crew duty and flight-time rules for each regime are versioned, independently testable rule sets. Per-carrier rule objects handle airline-specific overrides with no code changes.
Rolling-window FTL validator
An airline-agnostic validator checks duty and rest compliance across 24-hour, 7-day and 28-day rolling windows.
Memory with permissions
A namespaced memory layer built on Mem0 with lifecycle-managed context. It supports long-horizon multi-agent workflows with strict per-tenant isolation.
Vision into structure
Vision-language models turn crew rosters, NOTAMs and operational PDFs into schema-validated data for the rules engine and downstream agents.
TRY IT — LEGALITY LAB
Run a deterministic check yourself.
Move the sliders and the verdict updates instantly, with every rule shown. No model is involved. Thresholds are simplified for this demo and aren't regulatory guidance.
VERDICT
LEGAL
LLM tokens used for this verdict: 0
Three legs, one heading.
-
TRV PRESENT · in flight
Applied AI Engineer @ Torvalds
- Built a context-engineered system for airline operations automation, with multiple interacting agents and an upgraded permission-based Mem0 memory layer.
- Designed and engineered a legal validation system from the ground up, comparable to Jeppesen and AIMS, that verifies flight legality and regulatory compliance before takeoff.
- Architected an enterprise AI operations platform that replaces manual workflows with agent-based automation and vision-powered document processing.
- Deployed a multi-layer orchestrated workflow where tool-calling agents interface with airline APIs to retrieve live data, make updates and generate work orders for ground staff.
-
CLB 2023 — 2025
Software Engineer @ Cloudbloom
- Built and deployed GenAI backends on AWS Lambda, DynamoDB and Cognito, improving scalability and reliability across diverse AI workloads.
- Designed and optimised microservices for agent-based and LLM workflows using LangChain, LlamaIndex and LangGraph.
- Led development of event-driven AI pipelines for semantic search, conversational AI and real-time data processing.
- Worked in Agile teams to build and refine AI features in cloud-native environments as the GenAI strategy evolved.
-
MDA 2022
Software Engineer Intern @ MadAlgos
- Built a real-time collaborative coding IDE with live video chat over WebSockets to simulate a realistic interview environment.
- Designed Azure pipelines for continuous integration and coordinated feature deployments for EdTech products.
- Contributed to AI-driven UI/UX improvements and the backend for collaborative learning and code-assessment platforms.
-
IIIT ORIGIN
B.Tech, Computer Science @ IIIT Dharwad
Indian Institute of Information Technology Dharwad, Karnataka · GPA 7.27
★ Flipkart GRiD 2023 Semi-finalist ★ Runner-up, Inter-college Startup Hackathon
04 — BOARDING PASSES
More things I've shipped. Tap a pass for details.
RAGHTW
Hitwit
A GenAI learning and interaction platform: RAG chat, AI podcasts, auto-generated courses and mind maps.
CLASSEdTech · GenAI
STACKLambda · SQS · SES
SEAT01A
- Designed and built a RAG pipeline for chatbot conversations, podcast discussions and knowledge-base interactions.
- Integrated LLMs with Zoom and Slack to add assistant features to real-time collaboration.
- Built a dual-portal (admin/user) web platform for tracking, managing and analysing chatbot data in real time.
- Built content generators for podcast dialogue, auto-created courses and AI-generated mind maps.
- Backend on AWS Lambda, DynamoDB, SQS and SES for high availability and performance.
- Built a Chrome extension that summarises web pages and generates insights and visuals with contextual AI.
- Built in-site widgets for AI-powered guidance and navigation based on the site map.
HRTLF
TalentFlow
An agentic HR system: automated screening, resume parsing, smart assessments and AI-written candidate communication.
CLASSSaaS · Multi-tenant
STACKMicroservices · RBAC
SEAT02C
- Built a GenAI recruitment SaaS with automated candidate screening, resume parsing, smart assessments and AI-powered communication.
- Engineered an end-to-end microservices architecture with secure multi-tenant management, JWT authentication and granular RBAC.
- Built real-time dashboards using GenAI for performance prediction, skill matching and recruitment insights.
- Integrated AWS S3, LinkedIn, Gmail and Outlook APIs for document, search and communication workflows.
CHATF36
Flo360
Event management run from chat. Natural-language commands on WhatsApp or Telegram become bookings and invites.
CLASSAgentic workflow
STACKn8n · Docker · EC2
SEAT03F
- Engineered an agentic AI workflow using n8n and LLMs to interpret natural-language commands from chat platforms such as WhatsApp and Telegram.
- Automated event creation, booking and user invites through GenAI agents orchestrated in n8n.
- Implemented a full CI/CD pipeline with Docker and AWS EC2 for automated deployment of backend services.
LMSARV
Arivu
A multi-tenant LMS with a custom AI orchestration engine and RAG over large collections of enterprise documents.
CLASSEnterprise · LMS
STACKCognito · RAG
SEAT04B
- Built a custom AI orchestration engine that works with multiple LLM endpoints to process dynamic queries and generate insights.
- Architected scalable knowledge bases for large enterprise document collections, with custom RAG pipelines that retrieve and combine context from multiple documents.
- Engineered authentication (AWS Cognito) and authorisation microservices for secure, scalable access control.
- Built a multi-tenant frontend for scalable onboarding and role-based feature access.
- Added fine-grained access control through a role-management microservice, improving data separation across tenants.
The tools I fly with.
AI / ML & agent orchestration
- LangChain
- LangGraph
- LlamaIndex
- Multi-agent systems
- MCP servers
- Tool-calling frameworks
- RAG pipelines
- LLM fine-tuning
- Mem0
- Vision-language models
- Guardrails
- Hugging Face
Languages
- Python
- TypeScript
- JavaScript
- Go
- C / C++
- SQL
- HTML / CSS
Cloud & infra
- AWS Bedrock
- Lambda
- API Gateway
- DynamoDB
- SQS
- SES
- S3
- Cognito
- EC2
- IAM
- Amplify
- Azure
- Docker
- n8n
- Firebase
Backend
- FastAPI
- Flask
- Node.js
- Express
- Serverless
- Golang
Frontend
- React
- Next.js
- React Native
- Redux
- MUI
- PHP
Data & vector stores
- PostgreSQL
- pgvector
- Pinecone
- Qdrant
- MongoDB
- DynamoDB
- Aurora RDS
- MySQL
- SQLite
- BeautifulSoup
- Selenium
07 — FINAL CALL