Amit Maheshwari

AI Agents

Autonomous reasoning & collaboration — agents that reason, plan, and work together to solve complex problems end-to-end.

• Multi-Agent — Deep Thinking What happens when an AI summarizes a policy document and gets it wrong — and a customer acts on it?
  
  The fix isn't a longer prompt — it's a thinking pipeline.
  
  I built this with a multi-agent deep research summarizer. The architecture translates directly to customer support, compliance, and knowledge management.
• Multi-Agent — AI Story for Kids What happens when an AI generates the wrong answer for a credit card customer? Or hallucinates a banking term?
  
  The fix isn't a smarter model — it's a safer pipeline.
  
  I prototyped this pattern with a kids story generator. The architecture translates directly to banking, insurance, and fintech.
• Engineering Team Dashboard How many sprints does it take to go from a plain-English requirement to working, tested code — with design, backend, frontend, and unit tests all done?
  
  What if each handoff was an Agent?
  
  I built a 4-agent sequential pipeline: Design → Backend → Frontend → Tests. Each agent's output is the next agent's input — no human in the loop.
  
  The architecture mirrors how real engineering teams are structured. The same pattern applies to any multi-step enterprise workflow.

RAG Applications

Retrieval-Augmented Generation — combining information retrieval with GenAI to deliver accurate, context-aware responses.

• Evolution of the Internet Every AI application that answers from your data starts here.
  
  This is my "Hello World" of RAG — a chatbot grounded in a single document. Ask it anything about the Evolution of the Internet, and it answers only from that source. No hallucination.
  
  Before building enterprise RAG, you need to understand this.
• Image Insight — Pinecone Vector Store Most RAG demos skip the hard parts — security, embeddings, logging, encryption.
  
  This one doesn't.
  
  I built a production-grade RAG pipeline: CloudFront → API Gateway → Lambda → Amazon Bedrock (Titan Embeddings + Nova Pro) → Pinecone Vector DB. KMS encryption. CloudWatch logs. IAM policies.
  
  The architecture is ready to drop into any enterprise knowledge search use case.
• Image Insight — S3 Vector Store Any 3rd party vendor like Pinecone is good. But what if you're already on AWS and don't want another vendor?
  
  Same production-grade RAG pipeline — CloudFront → API Gateway → Lambda → Amazon Bedrock — but the vector store is S3. No external database. No extra cost. No new service to manage.
  
  Sometimes the best architecture is the one that uses what you already have.

DevOps Engineering

Productivity and delivery tooling — DORA metrics visualisation and an Eisenhower Matrix task manager, both deployed on Oracle Cloud with full auth and persistent storage.

• DORA Metrics Dashboard
• Eisenhower Task Manager Your delivery team has 40 backlog items — everything's marked "urgent." How does the PM decide what actually moves this sprint?
  
  The fix isn't a longer standup — it's forcing a 2×2 decision: urgency vs. importance.
  
  I built this Eisenhower Matrix app with React, Flask, Oracle ADB, and Firebase Auth. The prioritization model maps directly to sprint planning in fintech, banking, or any team running parallel delivery tracks.

Mortgage Platform Delivery

End-to-end delivery of DMT's enterprise mortgage technology platform across 10+ client onboardings. Led PI planning, established Agile governance, and owned offshore release management — delivering on time with zero critical post-deployment incidents.

AI Loan Officer Assistant

Building an AI-powered assistant (Co-Pilot + Empower LOS) to automate mortgage pipeline queries and accelerate loan officer decision-making. Also deployed an AI project assistant that reduced PM reporting effort by ~40%.

Jira → Azure DevOps Migration

Spearheaded an org-wide migration from Jira to Azure DevOps across 6+ global cross-functional squads — improving delivery visibility, traceability, and reporting standardisation for both technical and business stakeholders.