Chris Cordaro Applied AI · ML systems · Production

Applied AI · ML systems · Production

I build systems that ship, and I run the teams that keep them running.

Twelve years turning messy data and ill-defined problems into working software: a production RAG and multi-agent SaaS I architected and shipped solo, applied ML (forecasting, risk scoring, NLP) inside a federal agency, and a 400-user analytics platform I led the build of. Python first, shipped in containers, tested in CI.

Download resume See the work →

Proof

Outcomes, not decks.

  1. <2 min

    Opportunity research, down from ~2 hours

    A RAG copilot over 50,000+ federal solicitations: hybrid full-text and vector retrieval, intent classification, streamed answers with citations. Built solo, in production.

  2. ~6 min

    15-page intelligence reports, down from 8+ hours

    A four-phase multi-agent pipeline (research, analyze, write, review) with live progress the reader can watch.

  3. 30%

    Better workforce forecasts at a federal agency

    LSTM forecasting, XGBoost attrition scoring, and genetic-algorithm allocation at the DoD Inspector General. Manual analysis fell 60% and time-to-hire 10%.

  4. 400+

    Users on a self-service analytics platform

    A modern-data-stack migration over 40+ sprints: a 35% cost and scalability gain and 98% on-time delivery.

Recognition

  • UkraineOversight.gov, which I designed and delivered, was cited by CNN, C-SPAN, and Congress
  • Glenn/Roth Award for Exemplary Service
  • 30+ open agent tools, 150+ installs across three CLIs

Selected work

Build. Model. Lead.

Three stories: a product I built alone from schema to production, an ML platform I shipped inside a federal agency, and the team and platform I grew over six years.

  1. Build

    Scout: a production AI research platform, built solo from schema to ship

    Federal business development runs on manual searching, duplicated records, and days of research per pursuit. I built the platform that does it in minutes.

    The problem

    Opportunities surface across four government sources (SAM.gov, USASpending, FPDS, SBA) with conflicting metadata and heavy overlap. Analysts spent hours per pursuit re-reading solicitations and assembling competitive intelligence by hand.

    What I built

    A user searches or asks a question and gets a cited, streamed answer grounded in 50,000+ live solicitations, plus a full competitive-intelligence report on demand. Underneath: multi-source ingestion with four-tier deduplication and automated validation, an AI composite score blending structured signals with vector similarity, a RAG copilot with hybrid full-text and pgvector retrieval and intent classification, and a four-phase multi-agent pipeline that researches, analyzes, writes, and reviews. Scheduled jobs keep the corpus current.

    How it’s built

    Python / FastAPI · PostgreSQL + pgvector · Docker · Cloudflare Zero Trust, Workers, R2 · Claude and Gemini APIs · GitHub Actions, pytest, Ruff, mypy

  2. Model

    Shipping applied ML inside a federal agency

    The DoD Inspector General planned its workforce in spreadsheets. I led the build of the ML platform that replaced them, and the change program that got people to use it.

    The problem

    Workforce planning was reactive and manual. Forecasts were guesses, attrition surprised leadership, and allocation decisions had no model behind them. The harder problem was trust: analysts and leaders had to believe the numbers before they would act on them.

    What I built

    An enterprise workforce-allocation platform: LSTM networks for demand forecasting, XGBoost for attrition risk scoring, genetic algorithms for constrained allocation, and BERT-based NLP for position and skill classification. Delivered across 24 two-week sprints against a 300+ story backlog, with every deliverable traced to a mission outcome. Model output surfaced in self-service dashboards so program offices could answer their own questions.

    How it’s built

    Python · TensorFlow, scikit-learn, XGBoost · SQL · Tableau · Agile with requirements traceability

  3. Lead

    Growing the team and platform that outlast the project

    Six years on one federal program: a team that grew, a platform rebuilt underneath it, and an on-time record that held through both.

    The problem

    Sustaining a multi-year data program means growing the team, modernizing the platform underneath it, and keeping the on-time record while doing both.

    What I did

    Over six years with the DoD Inspector General I directed a 15-person team of data scientists, engineers, and BI analysts; led a modern-data-stack migration across 40+ sprints; built a self-service platform for 400+ users; and founded a Data Science Center of Excellence for methodology, governance, and mentorship. I designed and delivered UkraineOversight.gov, the public dashboards for oversight of federal Ukraine funding. Today I lead a 16-person team on a Military Health System contract while still shipping code: ML and analytics pipelines for pharmaceutical supply-chain analysis, and an automated reporting pipeline that recovers ~20 hours a month.

    How it’s built

    Tableau · Spark · Python, SQL · Agile at 98% on-time delivery

How I build

Most AI initiatives fail between the deck and the deployment.

I stay accountable across the whole arc, so the design is buildable and the build serves the goal.

  1. Understand the data and the user first

    Start with what the user is trying to decide, what the data can actually support, and what “adopted” would look like. If the case doesn’t hold, say so before writing code.

  2. Prototype the risky part

    Pick the simplest architecture that can be secured, operated, and explained, then build the piece most likely to fail first, whether that is retrieval quality, model accuracy, or throughput, so the plan rests on something that runs.

  3. Ship in increments, with the plumbing

    Tests, CI, containers, observability, and traceability from every increment to an outcome. I stay hands-on in the code and the data, so estimates are real and the team ships working software.

  4. Measure adoption

    The work is done when people use it and the numbers move: time saved, decisions made faster, work that no longer needs doing.

Experience

Twelve years across data, ML, and product.

Short on purpose. The case studies above carry the detail.

  1. 2025 – Present

    Knowesis Inc.

    AI Engineering Lead & Program Manager

    Architected and shipped Scout, a production RAG and multi-agent SaaS, solo, while leading a 16-person Military Health System contract.

  2. 2024 – 2025

    DoD Office of Inspector General

    Enterprise Business Innovation Manager

    Led the applied-ML workforce platform (LSTM, XGBoost, BERT) and the agency’s data-driven transformation.

  3. 2019 – 2024

    Knowesis Inc. — DoD Inspector General

    Technical Lead & Program Manager, Data Platform

    Directed a 15-person analytics team, the modern-data-stack migration, and UkraineOversight.gov.

  4. 2018 – 2019

    Knowesis Inc. — FEMA / Defense Health Agency

    Solution Architect

    Cross-platform disaster-recovery data system on AWS; NLP-driven document classification platform.

  5. 2013 – 2018

    Atkins (SNC-Lavalin) — FEMA

    Data Manager

    Enterprise data-validation frameworks behind $300M+ in FEMA funding, 0.05% error rate.

Download resume (PDF) →

Capabilities

Five ways I create value.

Grouped by the problem each one solves. The tools underneath change every year.

  1. Hands-on engineering

    Ship the thing myself, schema to deployment, so plans stay honest and teams have a working reference.

    Python, TypeScript, SQL · FastAPI, Next.js, React · Docker, Kubernetes · GitHub Actions, pytest, Ruff, mypy · AWS, Azure, GCP, Cloudflare

  2. Applied ML and GenAI

    Take a model from notebook to production and prove it moved a number.

    Forecasting (LSTM), risk scoring (XGBoost), NLP (BERT), optimization (genetic algorithms) · RAG and hybrid retrieval · embeddings, pgvector · multi-agent orchestration · MCP · Claude and Gemini APIs · evaluation and validation frameworks

  3. Data platforms

    Build the pipelines and platforms that analysts depend on, and make them self-service.

    ETL and multi-source ingestion · deduplication and validation · PostgreSQL, Spark, Databricks · Tableau, Power BI

  4. Solution architecture

    Design systems that survive contact with real data, security requirements, and live operations.

    Zero Trust · secure by design in regulated environments · cost and latency budgets · observability

  5. Delivery and leadership

    Run delivery from roadmap to release and grow the team that owns it.

    Agile (PMI-ACP) · requirements traceability · 98% on-time delivery · mentorship · centers of excellence · change management

Certifications PMI Agile Certified Practitioner (PMI-ACP) — 2022 · Tableau Certified Associate — 2019 · Data Science Professional Certificate, HarvardX — 2021 · DataWalk Technical Professional — 2022

About

Builder with an analyst’s engine.

Chris Cordaro

I take a messy dataset or an ill-defined problem, find the metrics that matter, and turn them into working software. I build things when I am not being paid to: this site, a daily AI briefing that researches, writes, and publishes itself, and a steady stream of agent tools that other engineers install.

I have also led. Six years directing a 15-person data team, an agency-wide transformation program, and now a Military Health System contract taught me that the hard problems in applied AI are rarely the models. They are trust, data quality, security, and change. I like those problems, and I think the best engineers on AI programs are the ones who can carry the pager and still explain the result to the people who fund it.

Raleigh, NC.

Read the daily briefing →Agent tools on GitHub →

Federal and defense

Active SECRET clearance. Twelve years in DoD, DHA, and FEMA environments. Fluent in FedRAMP, CMMC, Zero Trust, FAR/DFARS, and OTA acquisition.

Contact

Let’s talk about what you’re building.

Looking for applied AI and ML engineering roles where I can own a system end to end: data in, model, product out. Also happy to talk with anyone building ambitious AI who wants a second set of eyes on the architecture.

[email protected]
Download resume
LinkedInGitHub