HomeImpactAI SystemsPersonal ProjectsGenAIML ExperimentsCase StudiesExperienceSkillsTestimonialsPM StoriesContact
Resume ↗
AI Product Leadership

Priyansha Nanda

Building intelligent products that create real-world impact.

35%

Anomaly detection accuracy uplift

42%

AWS spend reduction

20%

Revenue contribution influence

100K+

Institutional users adopted AI

View Case StudiesDownload Resume
35% accuracy lift in anomaly detection~$198K annual AWS savings20% revenue influence within 9 months100K+ institutional users reached15+ AI/TPM hires scaled1.5M+ user platforms delivered
Priyansha Nanda — Professional headshot

Executive Snapshot

AI product leader delivering measurable outcomes in market surveillance, GenAI assistants, and FinOps efficiency.

Responsible AICross-functional leadershipScaled orgs

Introduction

Hear it from me

A 60-second walkthrough of who I am, what I've built, and what I'm looking for next.

Watch My 60-Second Intro

Who I am, what I build, and what drives me

Impact

Outcomes tied to accuracy, adoption, and cost efficiency

Evidence of product value across regulated finance, AI safety, and enterprise scale.

0%

Anomaly detection accuracy uplift

AI-powered surveillance features

0%

AWS spend reduction

~$198K annual savings

0%

Revenue contribution influence

Within 9 months

0K+

Institutional users adopted AI

Market surveillance users

0M+

Scale of AI-UX platforms

Users across Euroclear tools

0M+

SaaS products supported

Multi-tenant Azure apps

AI-Native Production

Real-time, production-grade AI systems I built

End-to-end architectures processing millions of events daily — live in regulated finance.

Real-Time Anomalous Trade Detection

PythonSageMakerKafkaSnowflakeDatadog
Production

End-to-end ML pipeline ingesting 2M+ daily trade events, scoring anomalies in real time, and surfacing ranked alerts to 100K+ surveillance analysts.

2M+ events/day

Throughput

<120ms

Latency P99

99.97%

Uptime

Data Flow Architecture

Trade Events
Kafka Stream
Feature Store
Anomaly Model
Confidence Scorer
Alert API
Analyst Dashboard
Source Streaming Processing ML Model Output

Multi-Agent Insider Trading Assistant

BedrockClaudeRAGLangChainOpenSearch
Production

Orchestrates 4 specialized agents — news sensitivity, FX correlation, trader pattern, and summary — with confidence gating and traceable citations across 8 languages.

4 specialized

Agents

8

Languages

<8s e2e

Response

Data Flow Architecture

User Query
Orchestrator
News Agent
FX Agent
Pattern Agent
RAG + Citations
Summary Response
Source Streaming Processing ML Model Output

Predictive Cloud Cost Engine

AWSDatabricksTableauLambdaSNS
Production

Forecasting engine predicting AWS spend 90 days ahead with automated governance alerts, ownership attribution, and remediation triggers.

90 days

Forecast window

+28%

Accuracy

~$198K/yr

Savings

Data Flow Architecture

Cost Explorer API
ETL Pipeline
Forecast Model
Anomaly Detector
Alert Engine
Governance Dashboard
Source Streaming Processing ML Model Output

Personal Builds

Creative side projects with product depth

Where strategy, engineering, and storytelling come together in public.

Strategic Fintech Product Concept

LSEG Coin USDT

A boardroom-ready product story that reframes tokenization as institutional market infrastructure, not just crypto hype.

Built a premium, single-page presentation app exploring a dual-token model (stablecoin + utility token) to unify equities, tokenized assets, and crypto on one regulated venue.

ReactTypeScriptViteTailwind CSS
  • 16-section executive narrative from market opportunity to implementation roadmap
  • Scenario modeling with base case projection to $407.5B annual revenue
  • Unified trading workflow and financial engine visuals for fast stakeholder alignment
Signature move: Reimagined stock-exchange capabilities through an innovative dual-token model that modernizes settlement, unlocks cross-asset liquidity, and expands institutional trading features.
View Repository ↗

AI Career Operating System

UpSkillOS

From resume chaos to interview confidence: an AI copilot that converts career ambiguity into weekly execution.

Designed and built a full-stack Next.js platform that diagnoses skill gaps, generates role-specific roadmaps, runs mock interview feedback loops, and simulates salary outcomes.

Next.js 14TypeScriptTailwindRadix/shadcnOpenAIRechartsZod
  • End-to-end product flows: resume analyzer, roadmap planner, mock interviews, salary simulator, and projects + badges
  • Production-grade API patterns with validation and basic rate-limiting under `src/app/api/*`
  • Runs with graceful fallback to high-quality mock responses when provider keys are unavailable
Signature move: Blended product strategy, UX storytelling, and AI capability design into one system users can actually execute week by week.
View Repository ↗

GenAI Feature Catalog

GenAI features shipped with retention & adoption

Every feature measured by weekly active usage, adoption scale, and business impact.

News Sensitivity Scoring

Live

LLM-powered classification of financial news articles by market-moving potential with explainable confidence bands.

LLMNLPClassification

Retention

87% weekly active

Adoption

12K analysts

Reduced manual news triage by 65%

Multilingual FX Insight Generator

Live

RAG pipeline generating contextual FX analysis summaries across 8 languages with source traceability and analyst override.

RAGMultilingualGenAI

Retention

82% weekly active

Adoption

8.5K users

3x faster alert enrichment

Trader Pattern Summarizer

Live

Multi-turn conversational agent that surfaces historical trade patterns and behavioral clusters for compliance investigations.

Multi-agentConversationalEmbeddings

Retention

79% weekly active

Adoption

6K investigators

+40% investigation insight depth

AI-Powered Alert Prioritizer

Live

Confidence-scored alert ranking using ensemble of rule-based and ML signals, with explainability layers and analyst feedback loops.

MLRankingExplainability

Retention

91% weekly active

Adoption

100K+ users

22% reduction in alert review time

Spend Anomaly Narrator

Live

GenAI narratives explaining cloud cost anomalies in plain English, with auto-generated remediation suggestions for engineering teams.

GenAIFinOpsSummarization

Retention

74% weekly active

Adoption

3K engineers

42% AWS spend reduction

OKR Progress Copilot

Live

AI assistant that auto-tracks OKR progress from Jira/Confluence, drafts status updates, and flags at-risk objectives with suggested actions.

GenAIProductivityRAG

Retention

85% weekly active

Adoption

20+ PMs & leads

20% improvement in delivery predictability

ML Experiments & Results

Experiments that moved the needle

Hypothesis, methodology, before/after metrics, and production outcomes — with evidence.

Click any experiment to see full methodology and before/after metrics

Hypothesis

Adding temporal features (rolling 7d/30d trade volume) and entity embeddings would improve precision without sacrificing recall.

Methodology

A/B test across 3 markets over 8 weeks. Shadow mode for 2 weeks, then 50/50 split with compliance sign-off.

Results

MetricBeforeAfterDelta
Precision68%84%+16%
Recall91%89%-2%
F1 Score0.780.86+0.08
False positive rate32%16%-16%

Production Outcome

Shipped to production. Net 35% accuracy uplift with 50% fewer false positives. Analyst trust scores improved from 6.2 to 8.4/10.

Hypothesis

Hybrid retrieval (BM25 + dense embeddings) with re-ranking would improve answer relevance for the Insider Trading Assistant.

Methodology

Offline eval on 500 curated Q&A pairs, then online A/B with 2K analysts over 4 weeks. MRR@10 and human relevance scores tracked.

Results

MetricBeforeAfterDelta
MRR@100.620.81+0.19
Answer relevance71%89%+18%
Hallucination rate8.5%2.1%-6.4%
Citation accuracy76%94%+18%

Production Outcome

Deployed hybrid retrieval with cross-encoder re-ranker. Hallucinations dropped 75%. Weekly retention jumped from 68% to 82%.

Hypothesis

Prophet with external regressors (team size, deployment frequency) would outperform ARIMA for 90-day spend forecasting.

Methodology

Backtested across 12 months of historical data. Compared ARIMA, Prophet, and LightGBM on MAPE and directional accuracy.

Results

MetricBeforeAfterDelta
MAPE18.2%11.4%-6.8%
Directional accuracy72%89%+17%
Alert precision61%83%+22%
Forecast horizon30 days90 days+60d

Production Outcome

Prophet + external regressors won. Extended forecast window from 30 to 90 days. Governance teams now act on 3-month trends.

Highlights

Flagship wins, built responsibly

High signal outcomes without confidential exposure.

AI Surveillance Accuracy

Shipped anomaly detection upgrades improving precision by 35% without sacrificing recall.

View AI Surveillance

Insider Trading Alert Assistant

Designed a multi-agent GenAI assistant for news sensitivity, trend plots, and multilingual FX.

View GenAI Assistant

Predictive Cloud Efficiency

Forecasting dashboards cut AWS spend by ~42% with actionable governance triggers.

View Cloud Efficiency

Case Studies

How I ship AI products at scale

Problem, approach, outcome, and learning — with evidence.

AI Surveillance: Anomalous Trade Detection

Improved surveillance accuracy and confidence scoring to flag high-risk trades early.

AI/MLMarket SurveillancePrecision/Recall
View case study →

Insider Trading Alert Assistant

Multi-agent GenAI assistant that assembles signals across news, trader activity, and FX.

GenAIMulti-agentResponsible AI
View case study →

Predictive Cloud Efficiency

Forecasting dashboards and governance triggers that reduced AWS spend by ~42%.

CloudFinOpsForecasting
View case study →

Operating Model

Systems that turn ambiguity into shipped outcomes

OKR governance, stakeholder councils, and prioritization rituals that scale.

OKR Engine

  • Quarterly goal setting tied to AI adoption
  • North-star metrics with guardrail KPIs
  • Clear ownership across product + data

Governance Cadence

  • Stakeholder councils with compliance, risk, and engineering
  • Decision logs for auditability
  • Prioritization rituals every sprint

Team Scale

  • Built Hyderabad AI/TPM org (15+ hires)
  • Managed 4 PMs, led 20+ person teams
  • Coached AI UX + data product practices

Delivery System

  • Roadmap cadence and release confidence scoring
  • Outcome reviews with leadership dashboards
  • Predictability improved by 20%

Experience

AI product delivery across regulated, enterprise scale

Click any role to expand full details

London Stock Exchange Group

Product Manager (AI/Data)

2023 — Present
  • 35% uplift in anomaly detection accuracy
  • Multi-agent Insider Trading Assistant
  • 42% AWS spend reduction (~$198K)

Full details

  • Led end-to-end product for AI-driven market surveillance across equities and FX
  • Designed multi-agent GenAI assistant with RAG, confidence scoring, and 8-language support
  • Built predictive cloud dashboards with governance triggers, saving ~$198K annually
  • Established OKR cadence and stakeholder councils across 20+ person org
  • Partnered with compliance to embed responsible AI guardrails and audit-ready logging

Capgemini (Euroclear)

Product Manager I

2022 — 2023
  • AI transaction tools +32% efficiency
  • 45+ releases led across 17-member team
  • AI-UX for 1.5M+ users

Full details

  • Owned product roadmap for AI-powered transaction processing tools
  • Drove 45+ feature releases across a 17-member cross-functional team
  • Improved operational efficiency by 32% through intelligent automation
  • Designed AI-UX patterns adopted across platforms serving 1.5M+ institutional users
  • Introduced prioritization frameworks that increased throughput by 45%

Capgemini (HSBC)

Associate PM

2020 — 2022
  • Fraud-risk dashboards cut resolution time 20%
  • Automation lifted delivery 37%
  • Bridged compliance, execs, and 4 eng teams

Full details

  • Built fraud-risk dashboards that reduced investigation resolution time by 20%
  • Introduced delivery automation that lifted team throughput by 37%
  • Served as bridge between compliance, executive stakeholders, and 4 engineering squads
  • Defined product requirements for regulatory reporting features
  • Drove adoption of agile rituals across distributed teams

Capgemini (Microsoft)

Software Engineer

2016 — 2020
  • Multi-tenant SaaS for 1M+ users
  • Azure provisioning automation at scale
  • Early product feedback loops on Azure ML

Full details

  • Engineered multi-tenant SaaS platform serving 1M+ enterprise users
  • Automated Azure resource provisioning, reducing deployment time significantly
  • Created early product feedback loops leveraging Azure ML capabilities
  • Contributed to architecture decisions for scalable cloud-native services
  • Transitioned from engineering to product thinking, informing career pivot to PM

Skills

AI product depth + execution breadth

End-to-end capabilities across strategy, data, and execution.

GenAI

LLMsTransformersRAGNLPAmazon BedrockSageMakerClaudeGPT-4.1Python

Cloud/Data

AWSAzureSnowflakeDatabricksSQLTableauPower BI

Product/Leadership

AI/ML roadmapsProduct strategyAI UXA/B testingPrecision/recallAdoption/retentionOKRsGovernancePrioritizationExecutive stakeholder managementMentoring/team building

Tools

FigmaJiraConfluenceDatadogMLflow

Testimonials

What colleagues and leaders say

Feedback from people I've worked with across engineering, data, and leadership.

“Priyansha has a rare ability to translate complex AI capabilities into clear product value. She drove our surveillance accuracy from good to exceptional.”

Engineering Director

LSEG

“One of the most effective PMs I've worked with. She built our operating model from scratch and aligned 20+ people behind shared outcomes.”

VP of Product

Capgemini

“Her blend of engineering depth and product intuition is unique. She doesn't just define requirements — she understands the technical trade-offs deeply.”

Senior Data Scientist

LSEG

“Priyansha redefined how we think about release quality. She introduced risk-based test strategies for AI features that cut our regression escapes by 40% while halving the test cycle.”

QA Director

LSEG

“Working with Priyansha felt different — she understood our CI/CD constraints, thought in terms of deployment blast radius, and always had an infra-aware plan B. Rare for a PM.”

DevOps Lead

LSEG

“She bridges the gap between architecture vision and product delivery like no one else. Her system design reviews are as sharp as any staff engineer's, but always grounded in user outcomes.”

Enterprise Architect

Capgemini

PM Storytelling

Real stories from the product trenches

Behind every metric is a decision. Behind every decision is a story worth telling.

Crisis → Framework

The 3 AM Pager That Changed My Prioritization Framework

A critical surveillance model started flagging 400% more alerts at 3 AM — all false positives. Instead of patching the threshold, I led a post-mortem that revealed we had no cost-of-error framework. I introduced a precision-recall cost matrix tied to regulatory penalty data. False positives dropped 60% in two sprints, and the framework became the default for every new model launch.

Lesson

“When the system screams, don't silence it — decode what it's telling you about your process gaps.”

A critical surveillance model started flagging 400% more alerts at 3 AM — all false positives. Instead of patching the threshold, I led a post-mortem that revealed we had no cost-of-error framework. I introduced a precision-recall cost matrix tied to regulatory penalty data. False positives dropped 60% in two sprints, and the framework became the default for every new model launch.

Click to read the full story →

Prioritization Courage

How I Killed My Own Feature to Save the Quarter

I had spent 6 weeks driving a GenAI summarization feature to beta. Usage data showed 12% adoption — far below threshold. Instead of doubling down, I presented the kill decision to leadership with a reallocation plan: redirect engineering capacity to the alert triage assistant, which had 4x the user pull signals. The triage assistant shipped 3 weeks later with 87% weekly retention.

Lesson

“A PM's job isn't to ship features — it's to ship outcomes. Killing your darling is a leadership act.”

I had spent 6 weeks driving a GenAI summarization feature to beta. Usage data showed 12% adoption — far below threshold. Instead of doubling down, I presented the kill decision to leadership with a reallocation plan: redirect engineering capacity to the alert triage assistant, which had 4x the user pull signals. The triage assistant shipped 3 weeks later with 87% weekly retention.

Click to read the full story →

Stakeholder Judo

Turning a 'No' From Compliance Into a Product Advantage

Compliance rejected our GenAI assistant because it couldn't explain its reasoning chain. Instead of fighting the 'no', I partnered with the compliance team to co-design an explainability layer — confidence scores, source citations, and override logging. The result wasn't just approval: it became our #1 differentiator. Analysts trusted it more than competing tools precisely because of the transparency.

Lesson

“Constraints aren't blockers — they're design briefs from your most demanding users.”

Compliance rejected our GenAI assistant because it couldn't explain its reasoning chain. Instead of fighting the 'no', I partnered with the compliance team to co-design an explainability layer — confidence scores, source citations, and override logging. The result wasn't just approval: it became our #1 differentiator. Analysts trusted it more than competing tools precisely because of the transparency.

Click to read the full story →

Data-Driven Influence

The Spreadsheet That Saved $198K in Cloud Spend

Engineering said cloud costs were 'just the cost of scale.' I built a single spreadsheet mapping every service to its utilization rate, overlaid with actual traffic patterns. The visual was undeniable: 42% of spend was on idle or over-provisioned resources. I presented it at the engineering all-hands with a 90-day rightsizing roadmap. Finance approved same day. The $198K savings funded our next ML hire.

Lesson

“You don't need authority to drive change — you need the right data in the right room.”

Engineering said cloud costs were 'just the cost of scale.' I built a single spreadsheet mapping every service to its utilization rate, overlaid with actual traffic patterns. The visual was undeniable: 42% of spend was on idle or over-provisioned resources. I presented it at the engineering all-hands with a 90-day rightsizing roadmap. Finance approved same day. The $198K savings funded our next ML hire.

Click to read the full story →

Proactive Leadership

Building an Operating Model Nobody Asked For

When I joined, there were 6 PMs, 20+ engineers, and zero shared rituals. Roadmaps lived in different tools, priorities conflicted weekly, and delivery predictability was at 55%. I didn't wait for permission — I drafted an operating model over a weekend: weekly syncs, a shared prioritization rubric, decision logs, and a lightweight OKR cadence. Within one quarter, delivery predictability hit 85%.

Lesson

“The best product you can build is sometimes the process that lets your team build better products.”

When I joined, there were 6 PMs, 20+ engineers, and zero shared rituals. Roadmaps lived in different tools, priorities conflicted weekly, and delivery predictability was at 55%. I didn't wait for permission — I drafted an operating model over a weekend: weekly syncs, a shared prioritization rubric, decision logs, and a lightweight OKR cadence. Within one quarter, delivery predictability hit 85%.

Click to read the full story →

Insights

Thinking out loud on AI product craft

Perspectives on building responsible AI products at enterprise scale.

AI/ML5 min read

Why Precision-Recall Trade-offs Matter More Than Accuracy in AI Surveillance

Most teams optimize for accuracy. In regulated finance, the cost of a missed alert far outweighs a false positive. Here's how I think about it.

Read insight →

GenAI7 min read

Building GenAI Assistants That Analysts Actually Trust

RAG pipelines and multi-agent architectures are powerful, but adoption depends on confidence scoring, override workflows, and traceable sources.

Read insight →

Leadership6 min read

The Product Operating Model That Scaled Our AI Org

OKRs alone aren't enough. Here's how stakeholder councils, decision logs, and lightweight rituals created real delivery predictability.

Read insight →

Beyond Work

What keeps me inspired

Creative and active pursuits that fuel fresh thinking.

Guitar

Playing guitar fuels creativity and helps unwind after intense product sprints.

Sketching

Sharpens visual thinking and design intuition — skills that carry into product UX.

Lawn Tennis

Competitive rallies build strategic thinking and quick decision-making under pressure.

Horse Riding

Requires focus, balance, and trust — much like leading cross-functional teams.

Contact

Let's build the next AI product platform

Reach out for AI product leadership, advisory, or roadmap acceleration.

LinkedInGitHubSend Email

Open to AI/data product leadership roles, advisory, and strategic builds.