Let's talk

PORTFOLIO

Hello, I'm

PriyeshDave

Senior AI EngineerEnterprise agentic systems

I architect enterprise GenAI and multi-agent platforms that take work off people's plates — from strategy and governance to production systems at American Express. 8+ years turning research into production.

Flagship work → Get in touch
Priyesh Dave 3× Patent inventor Agents in production
American ExpressBengaluru, IndiaFrom prompts to production
Multi-agent systems✦MCP✦A2A✦LangGraph✦GraphRAG✦Knowledge graphs✦LLMOps✦Fine-tuning✦Kubernetes✦GCP · AWS✦ Multi-agent systems✦MCP✦A2A✦LangGraph✦GraphRAG✦Knowledge graphs✦LLMOps✦Fine-tuning✦Kubernetes✦GCP · AWS✦

Flagship work

Work thatships.

Four programmes — from enterprise strategy to running systems. Each one pairs an architecture with the outcome it was built to move.

01American Express · Strategy

Agentic Enterprise Blueprint

Digital Workplace Data & Agentic AI Strategy

The enterprise roadmap for how agents are designed, governed and adopted across the Digital Workplace — for knowledge discovery and next-generation servicing.

  • Defined the strategy and roadmap that scaled agent adoption on ChatGPT and Microsoft Copilot.
  • Established reusable MCP and A2A interoperability patterns and simplified the Amex Agentic Brain architecture.
  • Authored ADRs that secured architecture, risk and governance approval for production agentic workflows.
  • Led the build-vs-buy assessment for incident categorisation — one of 21 critical TechCare journeys — across custom AI, ServiceNow Now Assist and Moveworks.
Agent strategyMCP · A2AADRs & governanceBuild vs buy
Reference architecture● live
Business journeys
Knowledge discoveryServicingIncident ops
Governed agent patterns
MCP toolsA2A handoffsADR guardrailsRisk review
Orchestration core
Amex Agentic BrainShared context
Adoption surfaces
ChatGPTMicrosoft Copilot
02American Express · Platform

Enterprise Knowledge Intelligence Fabric

Knowledge curation for ChatGPT & Copilot agents

A knowledge management layer that curates enterprise knowledge and serves it to ChatGPT and Microsoft Copilot — so agents answer from trusted, current sources rather than guesswork.

  • Designed and built the enterprise-wide Knowledge Management Service, with Generative BI for on-demand analytics in natural language.
  • Connected curated knowledge to ChatGPT and Copilot agents for grounded answers and continuous knowledge curation.
  • Semantic retrieval over knowledge graphs, vector search and GraphRAG, so answers carry relationships, not just matching text.
Knowledge graphsGraphRAGVector searchGenerative BI
Reference architecture● live
Enterprise sources
Knowledge articlesRunbooksProduct dataAnalytics
Curation & semantic layer
Knowledge graphVector indexGraphRAGFreshness checks
Knowledge services
Retrieval APIsGenerative BIMCP connectors
Consumed by
ChatGPT agentsCopilot agents
03American Express · Production

Autonomous Incident Intelligence Platform

Smart incident data management, run by agents

Evolved incident data management from a BERT classifier into a real-time, agent-driven GenAI platform that categorises, groups, enriches and traces root cause across 100+ Digital Workplace products.

  • Agents handle categorisation, grouping and enrichment in real time; knowledge-graph reasoning surfaces root cause.
  • Raised grouping accuracy by 20 points to ~98% and cut incident turnaround by 87%.
  • Turned multi-day reporting into under 3 minutes and released 5 FTEs to higher-value work.
Multi-agentGenAIKnowledge-graph RCAServiceNow
Reference architecture● live
Signal intake
ServiceNow incidentsReal-time stream
Agent crew
ClassifyGroupEnrichRoot-cause
Reasoning
Knowledge graphHistorical patterns
~98%grouping accuracy
−87%incident turnaround
<3 minreporting, from days
04Independent R&D · Open demo

Orbit

Agentic enterprise intelligence workspace

Brings together the fragmented information, conversations, tasks, meetings and project signals enterprise teams work with every day — and turns them into decisions and next steps.

  • Catch Up turns meetings into decisions, actions, risks and follow-ups.
  • Daily Plan merges calendar, messages, tasks and deadlines into one plan.
  • Readiness coordinates specialised agents to investigate project risks and dependencies.
Multi-agent orchestrationPersona-aware reasoningMCPCost & latency observability
View repository →
Architecture● demo
Enterprise tools
MeetingsEmailMessagesProjects
Orbit intelligence layer
ContextPersonasAgent orchestrationObservability
Experiences
Catch UpDaily PlanReadiness
Outcome
DecisionsPrioritiesNext steps

How I work

Not just code.Code that ships.

A model in a notebook is a guess. The hard part is giving intelligence the context, constraints and governance to do meaningful work — then proving it keeps working.

  • 01

    Start from the business outcome

    Every system begins with the decision it improves and the metric that proves it — turnaround, accuracy, hours returned.

  • 02

    Evaluate before you scale

    Test suites, LLM evals and synthetic data come first, so quality is measured, not assumed.

  • 03

    Design agents to be governed

    Clear tools, clear boundaries, ADRs and MCP / A2A patterns a regulated enterprise can approve.

  • 04

    Observe everything in production

    Latency, cost, drift and tool calls are tracked from day one — LangSmith, MLflow, Grafana.

Ship it.
Measure it.
Make sure it never fails quietly.

— The rule I build by

Technical depth

The fullstack of AI.

01Generative & agentic AI

Multi-agent systemsMCPA2ALangGraphLangChainLlamaIndexRAGHugging FaceLoRA / QLoRAQuantisation

02Knowledge & retrieval

Knowledge graphsGraph databasesGraphRAGVector searchPineconeElastic

03ML & NLP

Deep learningTransformersBERTNERClusteringPredictive modellingTime-series forecastingStatistical analysis

04LLMOps & MLOps

LangSmithMLflowKubeflowAirflowEvidentlyGrafanaDockerKubernetesCI/CD

05Languages & frameworks

PythonJavaSQLPyTorchTensorFlowKerasscikit-learnspaCy

06Data & cloud

PySparkKafkaHivePostgreSQLGoogle CloudAWS

Open-source lab

Built inpublic.

Autonomous research & publishing agent

BlogMind

Ranks arXiv, Hacker News and GitHub signals with an LLM judge, then researches, drafts, critiques and publishes behind a PR approval gate. 24 posts live.

LLM-as-judgeHuman-in-the-loop
 AI curriculum engine

SkillFlow

Turns structured learning roadmaps into continuous, human-reviewed technical content — from generation to automated publishing.

AI agentsAutomation
 MCP workday planner

FocusPilot

Uses Model Context Protocol to gather context from enterprise tools and generate prioritised work plans.

MCPAI agents

Experience

The path sofar.

Jun 2025 — Present

Senior AI Engineer I

American Express · Bengaluru

  • Defined the Digital Workplace Data & Agentic AI strategy; scaled agent adoption on ChatGPT and Microsoft Copilot.
  • Established MCP / A2A patterns, simplified the Amex Agentic Brain and authored ADRs for production agentic workflows.
  • Own the Autonomous Incident Intelligence Platform across 100+ products — ~98% grouping accuracy, 87% faster turnaround.
  • Engineered the Semantic Data Intelligence Framework: survey analytics automation from ~30% to 95%.
  • Lead Enterprise Functional Test Suites — LLM-generated end-to-end tests with knowledge-graph dependency resolution.
Nov 2022 — Jun 2025

Data Scientist

American Express · Bengaluru

  • Built the first ML-based ServiceNow incident categorisation system — 8 days of manual work saved per 10k incidents.
  • Architected reusable ConfigML frameworks for training, monitoring, explainability and governance, adopted across teams.
  • Designed the enterprise Knowledge Management Service with Generative BI for natural-language analytics.
2018 — 2022

Data Scientist

Tata Consultancy Services · Westpac account

  • Four years of applied ML and NLP for one of Australia's largest banks — where the move from research to production started.

Writing & teaching

Ideas inpublic.

@beyondtheprompts.ai

Zero to
Agentic.

A 105-day series from AI fundamentals to production agents — a dev.to post and an Instagram Reel every day.

Read on dev.toWatch the Reels
Medium · Essay

Engineering Enterprise-Grade Synthetic Data Generation

→ Medium · Essay

From Problem to Solution: How AI Can Redefine Knowledge Management

→ Medium · Essay

The New Era of API Testing — semantic, autonomous, LLM-driven

→ Medium · Field notes

What survives after the demo?

→

Patents

3× Patent Inventor

American Express Inventor Award

Certified

Claude Certified Professional Solutions Architect

Certified

Google Cloud Professional ML Engineer

Certified

TensorFlow Certified Developer

Contact

Let's buildtogether.

Ship it for real.

Open to conversations about enterprise AI, agentic platforms, speaking and collaboration.

Emailpriyeshdave6@gmail.com LinkedIn/in/priyeshdave21 GitHubPriyeshDave Medium@priyeshdave90 Dev.to@priyeshdave6 X@priyeshdave6 Instagram@beyondtheprompts.ai