About me

Forward deployed engineering · agentic systems · RAG & evals

I build & deploy production AI systems for broker-dealers.

I turn ambiguous customer problems into production AI systems: RAG architectures, agentic workflows and multi-agent orchestration, shipped under real enterprise constraints.

View LinkedIn Download CV Read Research

300+

Aircraft fleet supported with ML-driven ops optimisation at IndiGo

85%

LLM token spend cut per broker (80 to 90%) with targeted surveillance

60%

Fewer hallucinated outputs, measured on a labelled eval set

1,200

Agent trajectories in my open AdequacyBench benchmark

Forward deployed engineering Agentic & multi-agent systems RAG & semantic search LLM evaluation MCP AI security & compliance AI observability

I’m a Forward Deployed Engineer. I turn ambiguous customer problems into production AI systems: RAG architectures, agentic workflows and multi-agent orchestration, shipped under real enterprise constraints. I came up through ML delivery inside a 300+ aircraft airline and through enterprise consulting, so before models, architecture or tooling, my first question is always, “what decision are we improving?”

As Founding AI Engineer at Frensei Innovation Labs (ClawbackVault AI), I own end-to-end technical delivery of an AI-first compliance intelligence platform for financial-services brokers, from discovery through production deployment. That includes LLM pipelines that cut token spend 80 to 90% per broker, a Claude-based signal-detection engine with full evidence-trail audit logging for regulated decisions, and a security posture that clears CASA Tier 2 and Google OAuth Restricted Scope. Before that, at IndiGo, one of the world’s largest airlines, I delivered ML-driven optimisation across a 300+ aircraft fleet with ~20% operational cost savings.

My MSc in Business Analytics at UCD Michael Smurfit (2025 to 2026) ran alongside full-time engineering, on top of formal AI & ML training from IIT Kanpur. I’m an open-source author of MCP servers, evaluation tooling and PyPI packages, first author of two research preprints on production AI systems, and I write and host two podcasts. I’m open to Forward Deployed Engineer, Founding AI Engineer, Applied AI Engineer and AI consulting roles in Ireland, Europe and globally.

What I do

  • ML icon

    Forward Deployed AI Engineering

    Embedding with teams to take AI from zero to production: scoping the business problem, shipping the LLM system, and iterating with real users until it delivers measurable outcomes.

  • DS icon

    AI Systems & LLM Architecture

    RAG pipelines, semantic search, MCP servers and agentic workflows with LangGraph. Multi-agent orchestration with eval harnesses, built for performance, cost and reliability from day one.

  • BA icon

    AI Security & Compliance

    AES-256-GCM encryption, PII masking ahead of every LLM call, prompt-injection protection, OAuth 2.0, and GDPR-ready data handling. Privacy engineered in, not bolted on.

  • QD icon

    Business Translation & Product Strategy

    “What decision are we improving?” Technical discovery, roadmaps and pricing architecture, communicated as fluently to founders as to infrastructure engineers.

Currently building

Founding AI Engineer & Product Lead · Live

ClawbackVault AI

Compliance intelligence that protects broker commissions from silent client churn.

I own the end-to-end build at Frensei Innovation Labs: a targeted-surveillance architecture that replaced full-inbox scanning and cut LLM token spend per broker by 80 to 90%, a three-stage processing pipeline (header scan → PII-masked body fetch → signal classification) where email bodies never touch the database, and a four-tier churn-signal engine on Claude Sonnet 4.6 across 15+ behavioural categories.

  • Eval harnesses, structured-output checks and calibrated confidence thresholds in the production pipeline
  • Agentic compliance-research workflows with planning, execution, verification and feedback loops
  • AES-256-GCM encryption, PII masking before every LLM call, prompt-injection protection
  • Row-level multi-tenant isolation across EU-West & Asia-Pacific; Gmail API & Microsoft Graph over OAuth 2.0
Claude Sonnet 4.6 LangGraph Evals Supabase OAuth 2.0 CASA Tier 2

How ClawbackVault processes a signal

The live pipeline. Drag to explore. Email bodies are PII-masked before any LLM call and never touch the database; targeting only the relevant mail is what drives the cost reduction.

0% LLM cost reduction per broker (80 to 90%)
Inbox HeaderScan PII-MaskFetch Signal· Claude 4-TierChurn jane@acme.co → ●●●●

Certifications & recognition

  • Global Leadership Programme

    Awarded by the Dean of the UCD College of Business, 2026

  • UCD Advantage Award

    UCD Smurfit · Honor, 2026

  • Data Strategy

    Professional Certification

  • Google Project Management

    Professional Certificate

  • BCG Advanced Analytics & Data Science

    Virtual Experience Program

  • Data Visualization & Analytics

    Experience Program

  • Relational Data on Azure

    Microsoft

  • Google Business Intelligence

    Professional Certificate

Tools & stack

AI engineering

Agentic AI Multi-agent orchestration RAG Semantic search Vector databases MCP Tool & function calling Context & graph engineering

LLMs & evaluation

Claude OpenAI Open-weight models LangGraph Eval harnesses Hallucination detection Structured outputs LLMOps

Production & platform

Python TypeScript SQL FastAPI Docker PostgreSQL (RLS) Redis Qdrant Neo4j DuckDB Supabase OpenTelemetry

Cloud, security & compliance

GCP Azure Multi-tenant SaaS Hybrid & VPC deployment OAuth 2.0 AES-256-GCM PII masking GDPR CASA Tier 2

Testimonials

  • David Williams

    David Williams

    Advait has a good sense of understanding data and the requirements that come along with it, and also knows the ways that are going to be effective to solve the problem statement.

  • Jessica Joseph

    Jessica Joseph

    Advait was hired on a contract to handle the my business data and present the necessary insights about it and to recommend the solutions that would further help me in expanding in the market. His work is phenomenal and his subject knowledge presents a deeper understadning of data and the insights that can be presented to solve a particular problem.

Daniel lewis
quote icon

Daniel lewis

Richard was hired to create a corporate identity. We were very pleased with the work done. She has a lot of experience and is very concerned about the needs of client. Lorem ipsum dolor sit amet, ullamcous cididt consectetur adipiscing elit, seds do et eiusmod tempor incididunt ut laborels dolore magnarels alia.

Resume

Education

  1. University College Dublin - Michael Smurfit Graduate Business School, Dublin

    Master of Science - Business Analytics
    September 2025 - August 2026

    Focus on business optimization, simulation, predictive analytics, machine learning, and technology consulting. Honors: Global Leadership Programme (awarded by the Dean of the UCD College of Business) and the UCD Advantage Award, 2026.

  2. Indian Institute of Technology, Kanpur

    Post Graduate Professional Certification Program - Artificial Intelligence & Machine Learning
    October 2023 - September 2024

    Advanced training in AI and ML with emphasis on deep learning and practical applications across multiple use cases.

  3. Christ University, Bangalore

    Bachelor of Business Administration - Business Analytics and Data Science
    June 2020 - June 2023

    Strong foundation in business management, strategy, corporate finance, and data-driven decision-making.

  4. Ryan International School

    Business/Commerce
    May 2018 - June 2020

    Business and commerce track with early exposure to analytics and economics.

Experience

  1. Founding AI Engineer & Product Lead

    Frensei Innovation Labs (ClawbackVault AI), Dublin
    December 2025 - Present

    AI-first compliance intelligence platform for financial-services broker-dealers. I own the full-stack AI architecture, enterprise deployment and product strategy from 0 to scale.

    • Cut LLM token spend 80 to 90% per broker with a three-stage targeted-surveillance pipeline (header scan, PII-masked body fetch, LLM signal classification) that replaced full-inbox scanning.
    • Enabled regulated churn decisions across 15+ behavioural categories with a four-tier signal-detection engine on Claude Sonnet 4.6, tiered confidence scoring and evidence-trail audit logging.
    • Cut hallucinated and inconsistent outputs by ~60%, measured on an internal eval set of labelled broker threads, by embedding eval harnesses, structured-output verification and calibrated confidence thresholds into the production pipeline.
    • Automated compliance-research workflows with agentic systems built around planning, execution, verification and feedback loops, orchestrated with LangGraph and tool calling.
    • Hardened the platform to CASA Tier 2 and Google OAuth Restricted Scope: AES-256-GCM at rest, PII masking before every LLM call, prompt-injection protection, GDPR export and Supabase row-level tenant isolation across EU-West and Asia-Pacific.
    • Built and led a cross-functional team spanning AI engineering, cybersecurity and analytics, and own the four-phase technical roadmap and pricing-tier architecture with the Founder/CEO.

  2. Teaching Assistant - Digital Technologies in Business

    University College Dublin
    January 2026 - May 2026

    Postgraduate tutorials on AI in business, digital transformation and data-driven decision-making, alongside full-time engineering at Frensei. Bridged academic frameworks with industry use cases and guided 150+ students through projects, case studies, and presentations.

  3. Business Analytics & Machine Learning Mentor

    topmate.io
    January 2025 - May 2026

    1:1 coaching on LLM application development and technical positioning for professionals moving into AI engineering and business analytics roles.

  4. Business Analyst & ML Engineer

    IndiGo (InterGlobe Aviation Ltd), Gurugram
    June 2023 - August 2025

    ML-driven optimisation and decision intelligence across a 300+ aircraft fleet at one of the world’s largest airlines. Promoted into the ML engineering role within six months; owned the bridge between data science and executive decision-making.

    • Delivered ~20% operational cost savings fleet-wide by designing and deploying aircraft zero-fuel-weight prediction and on-time-performance cost models into production.
    • Improved cross-functional decision quality by ~40% with production ML and BI insight frameworks on GCP and Azure for operations teams and leadership.
    • Ran technical discovery with Business Technology and operations stakeholders, turning ambiguous business problems into ML requirements, a technical roadmap and deployed systems.
    • Reduced decision latency for operations leadership by re-architecting reporting logic and model parameters around the decisions each output drove.

  5. Business Development Analyst (Team Lead)

    CHRIST Consulting
    June 2021 - June 2022

    Led a 15-member analyst team on market-entry and growth-strategy engagements. Drove structured analysis and executive storytelling that contributed to a 25% increase in new business acquisition and ~15% improvement in renewals and upsells.

  6. Data Science & Business Analytics Internships

    IndiGo, LetsGrowMore, The Sparks Foundation, EntryLevel
    2020 - 2023

    Built early experience in data analysis, predictive modeling, and business analytics across multiple industry-focused projects and remote internships.

My skills

  • LLM Systems & Agentic AI (RAG, LangGraph, Claude, Multi-Agent Orchestration)
    90%
  • Forward Deployed Engineering (0→1 Production AI, Customer Embedding, AI Evaluation)
    88%
  • AI Security & Compliance (AES-256-GCM, PII Masking, OAuth 2.0, GDPR)
    82%
  • Business Translation & AI Product Strategy (Roadmaps, Pricing, Exec Alignment)
    85%
  • Analytics & ML Engineering (Python, SQL, GCP/Azure, Forecasting, BI)
    80%

Research

I publish open, reproducible research on production AI systems engineering: how AI systems are deployed, verified and overseen inside real organisations. Where a paper makes an empirical claim, the code and data ship with it.

What Agent Traces Cannot Tell You: Evidentiary Adequacy of Runtime Records for Agentic AI Oversight

Governance platforms now compile agent telemetry into regulator-facing compliance evidence. This paper builds the first benchmark that tests whether those records can actually support the findings of fact that oversight requires.

  • 1,200 agent trajectories
  • 36,296 steps
  • 6 determination families
  • 4 record conditions
Determinations resolved correctly

Pooled over 1,200 trajectories, by record condition

  1. Application action log 16.3%
  2. OpenTelemetry GenAI spans 16.3%
  3. Governance layer 28.75%
  4. Label-propagating substrate 100%

  1. Standard spans match a plain log

    OpenTelemetry GenAI spans produced the same outcomes as a plain action log on all 1,200 trajectories.

  2. Weak records fail silently

    On human intervention, the action log answers 52% of cases and is wrong in 27% of those.

  3. The relation binds, not the typing

    An oracle with perfect classification but no derivation relation still errs 47% of the time.

Read the paper AdequacyBench code

Forward Deployed Engineering: A Systems Engineering Perspective

A structured review of 133 academic and practitioner sources. The peer-reviewed literature contains no study of the role, so the paper treats vendor and client as one system of systems and maps forward deployed work onto ISO/IEC/IEEE 15288. Three structural properties follow.

  1. Requirements are discoverable only through operation.
  2. Verification evidence must be produced inside an environment the supplier does not control.
  3. Delivered architecture mirrors the delivery organisation across the firm boundary.
  1. 1Joint problem discovery
  2. 2Environment & data assessment
  3. 3Situated construction
  4. 4Evidence production
  5. 5Acceptance & transition
  6. 6Operation & adaptation

  • Adaptation loop returns stage 6 to stage 2, because the environment is what has usually changed.

  • Productization loop returns generalisable capability to the vendor’s product. Without it, the arrangement is consulting.

The cross-organizational deployment life cycle derived in the paper. Authority is divided at every stage.
Read the paper

Production AI Systems Engineering

  • Ongoing open research programme

The umbrella for everything above: treating the deployment and operation of AI systems as an engineering discipline with its own lifecycle, failure modes, measurements and standards. Planned and in-progress topics:

  • Context engineering
  • Enterprise agent architecture
  • Agent reliability
  • AI evaluation frameworks
  • AI observability & auditability
  • Enterprise RAG systems
  • Multi-agent coordination
  • AI technical debt
  • Re-qualification under model version churn
  • Multi-tenant isolation for agent platforms
  • The grounding cost curve
Open the research archive

Portfolio

A selection of what I’ve shipped, from a production AI fintech product and governed multi-agent platforms to MCP servers, open-source Python packages and applied ML. Explore more on GitHub.

Blog

Podcast

I host two shows: one decoding how business and technology converge, and one with bite-sized motivation for busy lives. Listen below or follow on Spotify.

The Business Technologist

Decoding digital transformation, AI & innovation for real enterprise value

  • From Data to Decisions — The Rise of Decision Intelligence
  • The Death of SaaS — And the Rise of AI-Native Enterprises
  • Hyperautomation: When Automation Becomes a Strategy
  • Agentic AI: Firing Your Chatbot, Hiring a Digital Worker
Open on Spotify

5 Minute Boosters

Bite-sized motivation & personal growth that fits your routine

  • Your Mind Is Lying to You
  • Your Future Self Is Watching
  • Success Isn’t Loud — It’s Consistent
  • Stop Apologizing for Who You Are
Open on Spotify

Contact

Open to Forward Deployed Engineer, Founding AI Engineer, Applied AI Engineer, and AI consulting roles in Ireland, Europe, and globally. If you’re building with LLMs, RAG, or agentic AI, the fastest way to reach me is email or LinkedIn, or drop a note below.

Send a message

Thanks for reaching out.

Your message has been sent successfully. I will get back to you soon.