Skip to content
Ayan Sasmal

Résumé

Ayan Sasmal

Lead Engineer · Platforms & AI

Lead Engineer, 13+ years in enterprise banking. Java and enterprise systems first, then cloud, then platform engineering, now AI platforms. Currently building agentic, LLM-assisted analytics that correlate activity across risk and compliance systems, after replacing a low-code post-trade platform with an in-house build and designing how AI agents get at internal data without anyone losing track of who authorised what. Most of my work comes back to the same thing: systems that stay accountable to the people responsible for them. Banking taught me to treat risk as a design input, not a review gate. Has led teams of 2 to 10 engineers, and now mentors more than 10 in parallel.

Technical skills

What I work with

AI & Agents
MCP serversAgentic platform designGoverned agent data accessRAG-as-a-Service integrationAI-assisted development
Backend
Node.jsTypeScriptPythonFastAPIExpressJava 7/8Spring BootHibernateTypeORMMongooseRESTful APIsSwagger / OAS
Frontend
ReactAngularSingle-spaMicro-frontendsStencilStorybookJavaScriptHTML5CSS3WebpackWCAG 2.1 AA
Platform & Cloud
AWS LambdaEC2S3DynamoDBRDSSNSSQSKubernetesDockerPostgreSQLMongoDBMySQLOracleDB2EnvoyOPAHashiCorp VaultEntra SSOJenkinsGitSonarQube
Architecture
MicroservicesMicro-frontendsEvent-drivenServerlessStrangler-fig migration12-factorSOLIDDRYSLAP
Risk & Assurance
Regulated-banking deliveryParallel-run migrationRollback planningAuditability by designLeast-privilege data accessOWASPSecure code by designFortify (SAST)Blackduck (SCA)OIDCSecrets management
Leadership
Team leadershipMentoringOKR alignmentTDDBDDCI/CDAgile (Scrum, Kanban)

Experience

Where I have worked

Macquarie Group

Sydney, NSW, AU · 2022–Present

  • Building the correlation engine and supervisor interface for a proof-of-concept analytics platform that brings activity from separate platforms, systems and channels into one trace, aiming to cut detect-and-report time from around 30 days to under five minutes.
  • Designing domain-specialised LLM agents: each evaluates every event against past activity and its domain’s regulatory definitions, explains events innocently first, then judges whether the series is a one-off or a deliberate act, citing its evidence.
  • Designed so a supervisor makes every decision: an AI recommendation with full reasoning, missing information they can add to trigger re-evaluation, and every decision recorded for audit.
  • Led the replacement of a low-code SaaS post-trade platform with an in-house Angular, Python and PostgreSQL system on Kubernetes.
  • Cut the heaviest form interactions from 30s+ to under 5s. The old screens fetched everything so users could search. The new ones query on two or three characters.
  • Built a faithful replica first, then optimised, running both systems in parallel to prove parity before cutover. That cost duplicated effort and meant Trading Ops never had to absorb new software and new behaviour on the same day.
  • Began decomposing the surviving internal API at the same time: off EC2 into containers, split into microservices by domain.
  • Led the first phase of a Single-spa micro-frontend shell (Angular, Python, Kubernetes) to host the division’s applications and its AI features, then handed it over to another team.
  • Designed and prototyped an MCP-based access layer for third-party AI agents: scoped, authorised and audited access to internal datasets the agents must never reach directly.
  • It replaced bespoke per-agent integrations and manual governance checks. Auditability stopped being something a team could forget to switch on.
  • Built the Agentic Solutions Platform. Event-driven ingestion from parquet, S3 and database sources on AWS serverless, normalised and served for scoped retrieval through another team’s internal RAG-as-a-Service.
  • Led teams of 2 to 6 engineers. Now mentoring more than 10 engineers in parallel, mostly one-to-one on design and review.
  • Onboarded a full team onto AI-assisted development tooling and practice.

FinXL Professional Services Pty Ltd

Sydney, NSW, AU · 2021–2022

client: Macquarie Group

Cognizant Technologies

Sydney, NSW, AU · 2020–2021

client: Westpac

Innovate IT Consultancy

Sydney, NSW, AU · 2020

Full Stack Developer

TechForce Services

Sydney, NSW, AU · 2019–2020

Infosys Technology Limited

India · 2012–2019

Senior Systems EngineerJan 2014 – Jan 2015

Systems EngineerApr 2012 – Jan 2014

Personal projects

Built outside work

Quorum

2025–Present

MCP · TypeScript · Open source (Elastic License 2.0)

  • Creator of an open-source governed memory layer for AI agents, served as an MCP server. It stores engineering decisions and business requirements with authorship and rationale, so agents read the reason behind the code, not just the outcome.
  • Conflict detection and authority weighting hold contradicting decisions before they land, and every change is recorded on a signed audit chain.

Education

Qualifications

Bachelor of Technology, Information Technology · GGSIP University, New Delhi

2007–2011