Michele Costa
Engineering Manager · New York

I build teams that make trusted data platforms.

Twenty years at Bloomberg leading the platforms that turn raw market data into decisions behind trillions in assets. Before that: CERN, Bell Labs, and a startup that got acquired.

Data platforms & pipelines Data quality & governance LLM & agent integration
Portrait of Michele Costa
Michele Costa, New York
i.

In Brief

My career started with particle physics and landed in portfolio analytics, which is less of a leap than it sounds: both are problems of moving enormous amounts of data quickly and reliably. At CERN I helped build the world's first distributed data-acquisition prototype for particle physics over high-speed networks. At a New Jersey startup called Xebeo, I was the first engineering hire; the platform demo I built helped secure a $30M funding round, and the company was later acquired.

Since 2006 I've been at Bloomberg, where I now manage engineering teams across New York, San Francisco and London. My teams own the systems that feed Bloomberg's buy-side analytics: an enterprise data lake, the ETL pipelines behind pre-computed portfolio analytics, and the data-quality platform that keeps all of it honest. The portfolio managers who rely on this data oversee trillions of dollars in aggregate.

The part of the job I care most about is the teams themselves. I've grown five technical leads into independent system owners, mentored engineers into managers, and built an organization across three sites that keeps its people. Lately I've been leading our push to put LLMs and agents to work inside enterprise data products, and shaping how a 500-engineer organization adopts AI.

ii.

Selected Work

01
Reliability at Scale

Teaching a distributed system to say no, politely

Bloomberg's portfolio analytics platform sits behind clients whose workloads arrive in unpredictable surges. Rather than buy our way out with hardware, I architected CFC, a credit-based flow control system that measures resource usage across an n-tier distributed system and uses historical statistics to regulate flows before they become outages.

CFC cut risk load by roughly 90% and held the platform steady through the onboarding of a new client consuming an additional 30% of bandwidth. The smooth rollout was the result of thorough pre-deployment analysis, which earned support from senior stakeholders. The platform also surfaced traffic patterns and client-usage data that now inform our SRE team, product managers, and the CFO office's commercial model.

02
Product Delivery

A data-quality product, from whiteboard to $2M

Data validation used to be something every client rebuilt in-house. I drove BDVS, Bloomberg's data-validation product, from concept to launch: a data-quality platform built from scratch in Python and Spark on Kubernetes, with a rule framework that cut maintenance operations by about 90% compared to what it replaced.

The product reached $2M in first-year sales, with T. Rowe Price among the clients validating it in production. Along the way my team introduced SLO-based contracts, mandatory data-readiness assessments, and behavior-driven development.

03
Data Platforms & AI

An enterprise data lake, and what LLMs should do with it

I led delivery of PEDL, Bloomberg's buy-side enterprise data lake, giving clients scalable SQL access over Trino and Iceberg with asynchronous notification when data is ready to query. We validated the direction directly with European enterprise clients and designed the platform around user journeys and self-serve access. On the pipeline side, my teams decreased latency 10x on the core service powering all on-the-fly ETL in portfolio analytics, using caching, a multi-level parallel retrieval framework, Apache Arrow, and Redis.

Now I'm defining what comes next: exposing these datasets to AI applications through MCP, so LLMs and agents can query financial enterprise data and analytics. As the representative of Bloomberg's 500-engineer buy-side organization, I helped shape the enterprise AI integration strategy and presented the data-lake and transformation architecture to senior management.

04
Cross-Org Execution

Shipping ESG analytics across six departments, on a deadline set by a summit

Integrating ESG data into Bloomberg's portfolio and risk analytics meant coordinating six departments with six different roadmaps, against a release date pinned to the Climate Summit. I led the integration end to end, and it shipped on schedule.

The platform this work fed into went on to win an ESG Investing Award for best ESG and sustainability analytics platform. Cross-department delivery is mostly unglamorous work: aligning data models, negotiating interfaces, and making sure that dependent systems scale. That is precisely why it is worth doing well.

iii.

Career

2021 — Present

Engineering Manager

Bloomberg L.P. · New York

Leading engineering across three sites for Bloomberg's buy-side enterprise suite: the PEDL data lake, BDVS data validation, PORT portfolio analytics pipelines, and precomputed analytics. Partnering with product, data science, sales, and the CFO office; driving LLM and agent adoption across the development lifecycle.

2006 — 2021

Software Engineer & Team Lead

Bloomberg L.P. · New York & Lugano

Led teams of up to eight engineers on portfolio analytics: integrated Barclays BRAIS with Bloomberg PORT behind a single unified interface, built an incremental attribution model for large-scale equity and fixed-income analytics, and shipped a post-trade amendment system spanning 60+ exchanges over FIX. Established CI/CD standards and performance-monitoring dashboards across multiple product lines.

Before Bloomberg

Earlier: Startups, Bell Labs, CERN

Xebeo / UTStarcom · Lucent Bell Labs · CERN

First engineering hire at Xebeo, where I built a highly scalable routing database, extended the UNIX kernel to run parallel TCP/IP stacks, and gave the investor demo that helped close a $30M round before the company's acquisition by UTStarcom. At Bell Labs, led three engineers on the line-card software for the first Gigabit router with extensive QoS. At CERN, designed hardware and software for the world's first distributed data-acquisition prototype for particle physics.

Published: “Results from an ATM-based Event Builder” ↗

Education

EPFL & University of Genoa

Lausanne, Switzerland · Genoa, Italy

Postgraduate course in Computer Networks at EPFL; MSc in Electronics Engineering from the University of Genoa. Published in IEEE publications; speaker at Bloomberg's Reliability Conference (RelCon).

iv.

Beyond the Job

Bloomberg volunteer team in hard hats in front of a Habitat for Humanity build site
With my Bloomberg colleagues on a Habitat for Humanity build. A different kind of infrastructure work.
Bloomberg volunteers packing food supplies at a City Harvest warehouse
Packing food with the team at City Harvest in New York.

Outside of shipping software, I mentor students through City Tutors, serve as a senior interviewer for junior hiring, and stay active in diversity recruitment. At Bloomberg I co-founded a company-wide Data Quality Meetup that grew past fifty members and is now part of the Reliability Guild.

I'm a dual U.S. and Italian citizen. I speak English, Italian, and Portuguese fluently, get by well in French. I am currently a CFA Level 1 candidate, on the principle that if my systems serve portfolio managers, I should understand their application.