Xiaobing (Peter) Shang
Federal Financial Manager · Data Scientist · Agentic AI Specialist · Digital Transformer
Senior federal financial professional with 18+ years of experience in financial management combining
AI/ML, data science, and full-stack development to build production systems at the intersection of policy and technology.
21 live production applications — click to explore
Professional Summary
Senior Federal Financial Management professional specializing in financial management, budgeting, enterprise analytics, AI integration, and internal controls. Extensive CFO-level advisory experience supporting $338B portfolios. GS-15 since 08/2020. Expertise spans financial governance, reconciliation, predictive analytics, risk analysis, technology modernization, enterprise data platforms, and AI-enabled decision-support solutions.
Core Competencies
Federal Financial Management & Governance
Financial Reporting & Compliance Oversight
Budget Formulation & Execution Strategy
Payroll & Workforce Cost Modeling
Machine Learning & AI-Enabled Analytics
Agentic AI Strategy & RAI Governance
Financial Systems & Data Integration
Congressional & OMB Financial Reporting
Advanced Data Analytics & Dashboards
Financial Statement & Ratio Analysis
Capital Budgeting & DCF Analysis
Quantitative Modeling & Forecasting
Professional Experience
Portfolio Manager (AI/ML RPA Automation, Analytics & Transformation)
FSMI, DCFO, Office of the Under Secretary of War (Comptroller), Department of War
Pentagon, DC | GS-0560-15
Lead the AI/ML, RPA automation, and Budget Analytics portfolios for the Comptroller. Direct three teams of 21 budget, data, and technical staff. Previously led the IGT Reconciliation, FSD, DARQ, and PPBE Reform portfolios.
- ▸Lead design and build of a federated agentic AI solution for audit remediation, targeting Working Capital Fund audit readiness in FY2027 and Department-wide readiness by December 2028
- ▸Evaluated 1,425 enterprise use cases across 14 material line items and 10 categories, prioritized by mission value, feasibility, data readiness, risk, and reuse to guide enterprise adoption and investment decisions
- ▸Established data validation, governance, integrity, and traceability controls for an enterprise user base of 30,000, enabling consistent financial reporting and cross-directorate analysis
- ▸Led high-visibility GAO engagements, coordinating data, analysis, evidence, and executive responses supporting oversight recommendations and DoD financial management improvement
- ▸Led and transitioned the IGT Reconciliation, FSD, and DARQ (Dormant Account Review – Quarterly) portfolios through key implementation; improved DARQ sampling and the review process
- ▸Directed integration of 20 general ledgers into centralized execution reporting covering $338B in obligations/expenditures and $200B in accounts receivable, payable, revenue, and expenses
- ▸Maintain and enhance the Department-wide Spend Plan Module comparing budget to execution and centralizing justifications for variances
- ▸Build agentic AI and ML prototypes on ADVANA and financial management data for anomaly detection, reconciliation analysis, and predictive forecasting
- ▸Developed the AI architecture framework and the MDAP placemat for senior leadership; delivered RPA automation for audit support
- ▸Wrote the first strategy statement for the Budget Analytics portfolio with business metrics and KPIs; led the analysis, action plan, and implementation of PPBE Reform Commission recommendations
- ▸Built payroll and workforce liability forecasting models and manpower tools tracking payroll cost, vacancy, retention, and fill rate
- ▸Mapped source systems, data fields, and posting logic for the Universe of Transactions (UoT) reconciliation and moved the design toward transaction-level traceability
Senior Budget Analyst
Program & Budget, Office of the Under Secretary of War (Comptroller), Department of War
Pentagon, DC | GS-0560-15
- ▸Managed formulation and execution review for national defense portfolios including Military Construction; issued Funds Authorization Documents
- ▸Reviewed 40 Budget Justification packages; issued 12 Program Budget Decisions; realigned and saved $2B+
- ▸Conducted financial trend and variance analysis; validated data behind 50+ Congressional testimony statements and OMB inquiry responses
- ▸Automated financial databases improving reporting speed and reconciliation accuracy
Financial Systems Analyst / Budget Analyst
Office of Inspector General, Department of Defense
Alexandria, VA | GS-0560-13 / GS-0505-14
- ▸Built automated spend plan forecasting and financial risk assessment models
- ▸Reconciled 700K+ transactions resolving $10M abnormal balances
- ▸Built the analysis that recovered $87M in threatened budget cuts
- ▸Corrected 180K transaction lines ($12M) from the FY19 retroactive pay raise
- ▸Performed COR duties overseeing contract cost, schedule, and performance compliance
- ▸Conducted internal control reviews and authored financial management procedures
Budget Analyst / Resource Manager / Program Analyst
U.S. Army (Multiple Commands)
Redstone Arsenal, AL | Germany | Korea | GS-11/12
- ▸Managed $45M payroll program for 500+ personnel; reduced payroll excess by 20%
- ▸Managed financial operation, internal control, GPC program and 1,100 individual GTC accounts; reduced GTC delinquency rate by 40%
- ▸Executed funds in GFEBS (SAP-based ERP) and conducted UMD/ULO reconciliations
- ▸Applied Earned Value Management to monitor acquisition cost, schedule, and performance
- ▸Supported $200M+ AMCOM programs; cost-benefit analysis achieved 20% cost and 30% man-year reduction
IT & Data Science
Analytics & Data
Languages & Libraries
Platforms & Tools
ML & Capabilities
Federal FM Systems
Technology & Systems
Full-Stack Development
LLM & Agentic Engineering
DevOps, Cloud & Domain
Education
M.S. Data Science
Saint Peter's University
M.S. Digital Forensics & Cyber Investigation
Univ. of Maryland Global Campus
M.S. Cybersecurity Technology
Univ. of Maryland Global Campus
Master of Business Administration
University of Phoenix
B.S. Accounting
Univ. of Maryland University College
B.S. Acupuncture
Shandong Uni of TCM
Training & Certifications
Certified Defense Financial Manager (CDFM)
DoD Financial Management Level III
DoD Financial Management Level II
DAWIA Financial Management Level II
IBM Data Science Professional Certificate
IBM Applied Data Science Specification
GWU Data Analytics & Visualization Boot Camp
CS50's Introduction to AI with Python
5-Day AI Agents Intensive Course
Foundry & AIP Builder Foundations
Deep Dive: Creating Your First Ontology
Speedrun: Data Science Fundamentals
Speedrun: Your First AIP Workflow
Portfolio & Projects
21 live production applications spanning federal financial management, AI engineering, and data science.
Knowledge · Skills · Abilities — Demonstrated Across Portfolio
Agent Build Tutor — Enterprise Agent Stack Playbook
Fifteen modules in dependency order, taught from a working local build · aisteps.vercel.app
A playbook for building an enterprise agent system and knowing why each piece comes when it does. Fifteen modules run from model serving to a self-evolving loop. Each one names its prerequisites, the stub to use if you build it early, and how the same idea looks on Databricks, IBM watsonx, Codex, Cursor, Claude Code, and a second machine. Lessons come from a working local reference build: a scheduled job sanitizes the project and publishes its changes, design notes, and retrieval eval results to the site.
15-Module Build Map
Model serving through a self-improving loop, grouped as Foundation, Contracts, Capabilities, The loop, Behaviors, and Systems
Dependency-Ordered Playbook
Each module states what it needs first and what to stub if you build it out of order, with a 'Why does A come before B?' comparison tool
Full Lessons With Code
Inference, RAG and knowledge graph, harness, and evaluation: 8 to 10 build steps each with Python code, a done-when check, and incident write-ups
6-Platform Crosswalk
All 15 modules mapped to Databricks, IBM watsonx, Codex, Cursor, Claude Code / Agent SDK, and a plain second machine (90 cells)
Sanitized Live Build Feed
A scheduled job on the build machine scans the reference build, redacts hostnames, paths, keys, and domain data, and publishes changes, design notes, and eval metrics
Ask the Tutor
Lesson-grounded Q&A answered by a local model when the build machine is reachable and by a cloud model otherwise
DataMatter — Department of War Budget Analytics
Authority, obligation, and outlay with full data lineage · datamatter.vercel.app
A budget analytics platform that answers three questions: where the money was authorized, where it was obligated, and whether the two can be traced to each other. Twenty-one USAspending, FPDS, J-book, AFR, and OIG datasets load through a Python and PostgreSQL pipeline where 709 control assertions gate every publish, and failed checks are disclosed as findings instead of suppressed. A retrieval-grounded AI analyst runs two locally trained models on a Mac Studio, ranks statute and the DoD FMR above secondary sources, and cites the source, vintage, and model behind every answer.
Authority → Obligation → Outlay
951 Treasury accounts and $2.22T in FY2026 budgetary resources traced from SBR (File A) through object class and program activity (File B)
709 Control Assertions
76 controls (SBR tie-out, File A/B, exhibit-to-J-book) run inside the load; a critical failure rolls back and the prior vintage stays live
Award-to-Account Recon
4.49M FPDS actions ($491.7B, FY2025) reconciled to File C; only 2.6% carry a TAS link, reported as linkage, never as missing money
Funds Control & Lapse
TAS-level obligation and outlay rates, $855.3B unobligated balance, and lapse exposure limited to expiring availability
Ask the Corpus
Two models trained on J-books and the DoD FM knowledge bank, served from a Mac Studio; authority-ranked retrieval; every answer names sources, vintage, and model
Research: Where the Time Goes
CR delay median −1.6 days; September contract share rose 14.8% → 18.8%; only 1 of 26 material weaknesses is testable on public data
LLM Powerhouse — Home AI Server Bridge
Public gateway to a local Mac Studio running Ollama · llmpowerhouse.vercel.app
A Next.js site that bridges to a real home AI server — a Mac Studio (M3 Ultra) running Ollama with three local models — over Tailscale Funnel and a custom Python gateway. Hardened for public exposure with a dependency-free reverse proxy, a live health-check dashboard, per-app API keys that revoke instantly, and one-click Start/Stop/Restart desktop controls for the physical machine — reusable infrastructure now onboarding 18+ other apps.
Home AI Server Bridge
Next.js site bridges to a Mac Studio (M3 Ultra) running Ollama with 3 local models over Tailscale Funnel
Zero-Trust Gateway
gateway.py: dependency-free Python reverse proxy in front of Ollama — the only thing making public exposure safe
170–180s Timeout Fixes
Root-caused 3 stacked timeouts (gateway/env/Vercel) that killed slow model replies before they could finish
Live Health + Postgres
/server-status page with live health checks and a persisted attempt log, replacing a single opaque offline badge
Per-App API Key System
manage_keys.py: add/rotate/revoke, hot-reloaded by file mtime — cut off any one of the other 18+ apps instantly, zero downtime
Chain-of-Thought UI
Extracts reasoning_content/thinking tags into a collapsible reasoning panel in the Live Demo chat
Budget Matter — How the Federal Budget Works
Follow every dollar, policy to public impact · budgetmatter.ai.studio
A federal budget education platform: a 22-step interactive process map, a legislative framework tracing constitutional and OMB authority, and a searchable 224-term DoD FMR glossary. Rebuilt from a legacy site into a production React, TypeScript, and Tailwind CSS application.
22-Step Process Map
Interactive budget lifecycle from strategic planning through congressional action, execution, and closeout
Legislative Framework
Constitution, appropriations law, and OMB guidance mapped to the purpose/time/amount test
15 Applied Projects
Case studies connecting a public-finance problem, method, capability, and decision outcome
224-Term DoD FMR Glossary
Full January 2024 DoD 7000.14-R glossary rebuilt as a searchable A–Z reference with source-page citations
OMB A-11 Section 20 Reference
Visual, searchable federal budget vocabulary and budget-authority concepts
React + Vite + Tailwind Rebuild
Legacy HTML/CSS/JS site modernized into a production TypeScript app, deployed on Google AI Studio → Cloud Run
brainbank — The AI Knowledge Bank
GitHub-synced knowledge pipeline · aibrainbank.vercel.app
A GitHub-synced knowledge base where an Obsidian vault auto-parses into markdown, tags, and a bi-directional link graph, with media offloaded to Cloudflare R2. A shared API ingests markdown, PDF text, and video transcripts, turning each capture into an AI-generated learning map.
GitHub-Synced Ingestion
Obsidian vault syncs through a GitHub-backed pipeline — markdown, frontmatter, and tags parsed and published automatically
Cloudflare R2 Object Storage
Images and attachments offloaded to R2, decoupling media storage from the app database as the vault grows
Bi-Directional Link Graph
Cross-references parsed from every capture and rendered into an explorable, continuously-expanding knowledge graph
Typed API, Multi-Source
Shared ingestion API accepts Obsidian markdown, PDF text, and YouTube transcripts through one pipeline
AI-Generated Learning Maps
Each capture is auto-structured into What/How/Why with an AI-built learning map and hands-on steps
Auth + Bilingual Publishing
Sign-in-gated capture with a published/draft workflow, served bilingually in English and 中文
AI Use Case Library — Federal FM, Audit & Finance
6,442 searchable AI use cases across 5 workbooks · aiinaction.vercel.app
A searchable library indexing 6,442 federal and DoD financial-management AI use cases across four source catalogs, including two 2025 OMB government-wide AI inventories. An insights dashboard scores each entry by priority, risk, and evidence type.
6,442 Use Cases Indexed
One searchable index across five source workbooks — DoD FM, Federal/Audit/Finance, and both 2025 OMB inventories
4 Source Workbooks
DoD FM Catalog v2.1 (1,381), Federal/Audit/Finance (550), OMB Individually Reported (3,611), OMB Consolidated COTS (900)
Priority & Risk Scoring
Every DoD FM use case scored against FY27/FY28 audit priorities, material line items, and portfolio tiers
4 Strategy Papers
DoD FM AI Integration Strategy, General AI Adoption, AI Best Practices, and Advana-FM Architecture Blueprint — Word + PDF downloads
Full-Text Search & Browse
Keyword search, workbook/mission-area filters, and a full evidence trail on every catalog entry
Insights Dashboard
Breakdown by evidence type, priority, and risk — flags self-reported vs. source-derived data before it reaches leadership
ANY FED — Federal Financial Management Portal
Any-agency federal FM intelligence · anyfeddemo.vercel.app
A federal financial management portal supporting all 28 CFO Act and independent agencies, spanning budget formulation through audit remediation. Backed by an AI and ML workbench and a three-provider AI FM Analyst chain across Gemini, Groq, and Claude.
28-Agency FM Portal
Any-agency switching — DoD & SEC folder data + 26 agencies via live USAspending fallback
Full FM Lifecycle
Budget, Accounting, Audit, Finance Ops, Internal Controls, Contracts & Acquisition modules
AI / ML Workbench
Models run locally on selected agency data — forecasting, anomaly detection, classification
AI FM Analyst
Gemini → Groq → Claude 3-provider chain grounded on the active agency's data slice
Document Analysis
J-book xlsx exhibits, FY2025 AFR + DODIG-2026-032 PDFs parsed into audit solutions
Data Provenance Layer
Folder source data · USAspending API · Treasury FiscalData · DuckDB lakehouse — lineage on every page
CogniBloom — AI Learning Companion
K-12 AI tutor, notes & gamified growth · cognibloom.vercel.app
A personal AI tutor for K-12 learning that combines subject-aware AI chat, structured study notes, and AI-generated quizzes and flashcards. Includes a daily planner, a financial-literacy journaling track, and a public learning chronicle, wrapped in a gamified progression system with levels and badges.
Personal AI Tutor
K-12 tutoring sessions with subject-aware AI chat — 44+ AI chats logged
Notes, Quizzes, Flashcards
29 structured study notes (AMC 8 / Mathcounts) with AI-generated quizzes and decks
Gamified Progression
XP, levels, and badges earned through consistent daily use
Learning Chronicle
Published daily-reflection diary entries with titled, narrated learning recaps
Investment Reflection
Daily financial-literacy journaling alongside academic notes
Daily Planner + Kid-Safe
Planner view layered on curated content, with safe AI boundaries and a distraction-free dashboard
DoD FM System Architecture Blueprints
17 DoD financial system deep-dives + research appendix · dodfmsys.vercel.app
An interactive reference covering 17 DoD financial management systems, including GFEBS, Navy ERP, DAI, and DEAMS, mapped through a consistent source-to-statement lineage model. Paired with a 12-section research paper on the full transaction lifecycle.
17 System Blueprints
GFEBS, GCSS-Army, LMP, GCSS-MC, STARS, SABRS, FAMIS, DDRS, Navy ERP, DAI, DEAMS, GAFS, CEFMS, DLA EBS, ABSS & more
Source→Detail→GL→Statement
Consistent lineage explorer, audit lens, and static reference toggle across every system page
12-Section Research Appendix
Executive summary through GTAS/DDRS reporting mechanics, JV lifecycle, and budget-to-disbursement systems
Audit Readiness Lens
Five assertions, FIAR/NFR/CAP remediation, and a worked audit example (Appendix B)
UoT Lineage Model
Source-to-statement traceability mapped consistently across all 17 systems
7 Worked Use Cases
Travel, contracts, inventory, property, leases, IGT, and the F-35 program, feeder-to-statement
FedFMMatter — Federal FM AI Analyst
Production AI for federal financial management · fedfm.vercel.app
A production AI analyst platform for federal financial management, covering budget and programs, audit and assurance, accounting and execution, and contracts and acquisition. Combines RAG-powered Q&A with pgvector semantic search and a multi-model AI waterfall for high uptime.
Budget & Programs
PB justifications, appropriations, mandatory/discretionary splits
Audit & Assurance
Findings tracker, CAP milestones, FY2028 audit readiness
Accounting & Execution
Obligation rates, ULO analysis, TAFS reconciliation
Contracts & Acquisition
FAR/DFARS compliance, EVM variance, cost proposal review
RAG Knowledge Base
pgvector semantic search with citation-backed AI responses
4-Model AI Waterfall
Gemini 2.5 → Groq Llama 3.3 → Llama 3.1 → Gemini 2.0
AXIOM — AI Engineering Sandbox
Production AI patterns & lifecycle · fedaxiom.vercel.app
An AI engineering sandbox built around a 10-stage production lifecycle, from goal definition through deployment and monitoring. Distills agentic patterns — ReAct loops, RAG pipelines, multi-agent orchestration, evaluation, and guardrails — into hands-on interactive tooling.
10-Stage AI Lifecycle
Define Goal → Deploy & Scale — full interactive pipeline
AI Studio
Hands-on model chains and production pattern experiments
Code Intelligence
Code analysis tooling for AI engineering workflows
Live Feed
Real-time AI domain content aggregation
Knowledge Base
AI/ML reference covering RAG, agents, evaluation, guardrails
Career & Notebook
Structured AI engineering learning tracker and notes
Storyline — Big Ideas for Curious Kids
5-story bilingual library, ages 7–14 · storyonwho.netlify.app
A bilingual interactive story library for kids ages 7 to 14, teaching behavioral-finance concepts such as second-level thinking and assets versus liabilities through narrated, illustrated fiction. Most stories are readable in both English and Chinese.
5-Story Growing Library
Ages 7–11: Leo's Five Secrets, Brain Backpacks. Ages 12–14: Outsmart the Crowd, The Magic Snowball, Escape the Rat Race — new titles ship regularly
Bilingual EN + 中文
4 of 5 stories readable in English or Chinese from the same page
Behavioral Finance for Kids
Three ages 12–14 stories teach second-level thinking, the psychology of wealth, and assets vs. liabilities through story, not lecture
Listen Along
OpenAI narration when configured, with device text-to-speech as a no-account fallback
Private by Default
Reading progress stored locally first; anonymous Netlify DB sync is optional
Cloudinary Media Pipeline
Every story illustration fetched, resized, and optimized for the reader's screen
The Study-Hour Ledger
ROI calculator + read-along audiobook, 2 pages · uselessgadget.vercel.app
An interactive ROI calculator that models a study hour as a compounding investment, with five adjustable levers and a year-by-year compound-growth chart. A second page extends the concept into a read-along audiobook with synced highlighting, a narrator picker, and adjustable playback speed.
Interactive ROI Calculator
5 adjustable levers compute hours of future free time returned by 1 hour of study today
Compound Growth Chart
Year-by-year visualization of value compounding from today to the payoff year
Research-Grounded Model
Built on Heckman's 7 to 13 percent annual return findings for early childhood learning investment
5 Tunable Levers
Compounding rate, years to payoff, hourly cost basis, future hour value, efficiency bonus
Read-Along Audiobook
Second page, "The Level-Up Rule": browser text-to-speech narration with synced highlighting, narrator picker, chapter navigation, and speed control
Applied Behavioral Design
Concept validated through rapid build-measure-iterate cycles with a real user
LearnAnything AI — Study Plan Engine
Multi-model AI learning planner · learnuseless.vercel.app
An AI study-plan generator that turns any learning goal into a 90-day plan with daily actions, quizzes, and gap analysis. Offers a 10-model dropdown across the Gemini, Groq, and Claude families, with provider-aware chaining, saved plans, and progress tracking.
Planner Agent
Turns any goal into a 90-day plan with daily actions, projects, quizzes, and checkpoints
10-Model Dropdown
Gemini 3.5/3.1/2.5, Groq Compound + 3 Llama, Claude Sonnet/Opus/Haiku — user-selectable
Multi-Model Chains
Provider-aware chaining with local fallback when no API/database is configured
Saved Plans & Tracking
Persistent plans with progress tracking, weekly milestones, and resume-where-you-left-off
Gap Analysis
Planner identifies skill gaps against the goal and adjusts the weekly load (10h default)
Plan → Study → Track
Single workflow from goal entry to readable plan to daily execution console
PeterClaude
Claude API Chat Interface · peterclaude.vercel.app
A custom AI chat application built on the Anthropic Claude API, with session-aware conversation management and context window optimization. Uses secure server-side API routing through Vercel serverless functions.
Claude API Integration
Direct Anthropic API with multi-turn dialogue and session history
Context Management
Session-aware history with context window optimization
Secure API Routing
Server-side key protection via Vercel serverless functions
Low-Latency Design
Real-time message delivery with typing indicators
TypeScript-First
Typed responses, error handling, production-grade patterns
Clean Chat UI
Responsive interface with message formatting and timestamps
ML & AI Knowledge Hub
Interactive ML algorithm reference · mlaithing.vercel.app
An interactive reference covering eight-plus machine learning algorithms with evaluation-metric comparisons and structured model-benchmarking guidance. Includes a production AI agent architecture reference emphasizing interpretability, scalability, and governance.
8+ ML Algorithms
Logistic Regression, SVM, Random Forest, SGD, KNN, Naive Bayes, K-Means, GMM
Evaluation Metrics
Accuracy, Precision, Recall, F1, AUC-ROC, Confusion Matrix
Clustering Techniques
K-Means and GMM with real-world DoD/federal use cases
AI Agents Framework
Production ReAct pattern with tool calling and agentic workflows
Working Examples
Complete Python/scikit-learn implementations with best practices
Interactive Design
Next.js 15 UI with code highlighting and algorithm comparisons
Palantir Learning Platform
Self-evolving Foundry/AIP knowledge base · palantirthing.vercel.app
A self-evolving knowledge platform covering 11 Palantir Foundry and AIP platform areas, with an automated Python news-scraping pipeline and AI-generated executive intelligence briefings.
11 Platform Sections
Foundry, Ontology, AIP, Apollo, Workshop, Quiver, Slate, and more
Gemini 2.5 + Search
Agentic AI with live Google Search grounding across all sections
Automated News Scraping
Python 3.12 pipelines via GitHub Actions for daily Palantir news
AI Executive Briefings
Geopolitical and contract signal analysis with automated summaries
Persistent Storage
Neon PostgreSQL + Prisma ORM for notes and knowledge pages
Self-Evolving Platform
Continuously updated knowledge base with AI-driven content discovery
MyThing — Personal AI Platform
Secure knowledge management & AI agents · shangthing.vercel.app
A secure personal knowledge-management platform with authenticated notes, structured tagging, and four specialized AI agents, plus automated tech-news aggregation. Built with 102 files and 6,300-plus lines of TypeScript.
Gemini 2.5 AI
AI chat with full context, live search, and RAG over notes
Tech News Feed
Python scraper + GitHub Actions ingesting RSS feeds daily
Daily Notes + AI
Journal with AI-generated headline, themes, and sentiment
NextAuth v5 Auth
Google OAuth + passphrase with private/public route protection
Fed Finance Q&A
AI knowledge base over OMB circulars and DoD policy
Full-Stack Platform
Next.js 15, React 19, Prisma ORM, Neon PostgreSQL, Vercel
AI Governance Intelligence
Multi-agent AI policy platform · aimlgov.vercel.app
A multi-agent AI console for risk analysis, compliance briefing, and policy research, drawing on live news intelligence from 25-plus regulatory and research sources. Includes RAG-powered chat over a document library and a tool that converts documents into visual risk diagrams.
Multi-Agent AI Console
Risk analysis, compliance briefing, and policy research agents
Live News Intelligence
25+ sources scraped every 6h — NIST, EU Digital, FTC, arXiv
PDF Notes & Knowledge
Upload docs, auto-extract text, generate AI summaries on Vercel CDN
RAG-Powered Chat
pgvector semantic Q&A over document library with citation context
AI Visual Aid Generator
Converts documents into risk matrices, timelines, and diagrams
3-Model AI Fallback
Gemini 2.5 Flash → Llama 3.3 70B → Llama 3.1 8B for 99%+ uptime
aimatter — AI Studio Process-Flow Compiler
Text-to-flow-graph AOT compiler · aimatter.ai.studio
A compiler that converts pasted technical text into structured JSON process nodes, rendered as lightweight cubic Bezier flow graphs. Built on React and TypeScript with an Express ingestion server and a Gemini-based compilation pipeline, with bilingual, locally saved workspaces.
AOT Text→Graph Compiler
Converts copy-pasted technical text into standardized JSON process nodes and edge connections
Cubic Bezier Flow Graphs
Low-overhead, eye-friendly rendering with multi-view routing — no heavy graphics-library footprint
React 19 + TS 5.8 Core
Core framework setup for the compiler UI and multi-view routing
Express CJS Ingestion
Server ingestion engine bundled with esbuild
Gemini 3.5 Flash Target
Target AI architecture powering the compilation pipeline
Bilingual localStorage State
User workspaces saved to localStorage on every edit, with one-click Export Backup portability
Budget Matter — DoD Analytics Portfolio (Original)
The project that started it all · budgetmatter.github.io
The original federal financial analytics portfolio — Tableau dashboards tracking $338B in DoD obligations, Python and VBA automation tools, scikit-learn workforce models, and a USASpending.gov ETL pipeline, all on a modular static site. Later rebuilt from the ground up as the React/TypeScript Budget Matter above; this legacy build stays live as the original record.
Tableau Federal Dashboards
Interactive dashboards tracking $338B DoD obligations, spend plans, and budget execution variance
VBA + Python Automation
Excel VBA tools and Python/Pandas scripts automating repetitive financial report transforms
ML Workforce Models
scikit-learn regression models for manpower/payroll forecasting, plus applied happiness-index case studies
D3.js + Leaflet Visualization
Custom geographic and temporal data-exploration dashboards built without a heavy charting library
USASpending.gov ETL Pipeline
SQL + Python pipeline extracting and transforming federal spending data for analysis
The Original Build
Peter's first public engineering project — still live as the historical record, later rebuilt as the React app above
Federal Financial Management
FY2027 DoD Budget Inside Analysis
FeaturedPortfolio-manager reference for the $1.45T President's Budget request — topline bimodal architecture ($1.1T discretionary + $350B mandatory), procurement deep-dive across 19 appropriation titles, MAC munitions 18-program analysis, AI Arsenal ($58.5B) vs Drone Dominance ($53.6B), Advana→War Data Platform trifurcation, and execution risk register. Built inside FedFMMatter as a domain-native analytical product.
DoD Audit Readiness Intelligence System
FeaturedFY2025 DODIG-2026-032 audit decoded: all 26 material weaknesses mapped by remediation theme (Systems/IT, Asset Valuation, Liabilities, Revenue/Budget, Oversight, High-Value Programs), Advana capability matrix covering 17 of 26 MWs, dollar-at-risk log-scale analysis ($859B unsupported adjustments), FY2028 clean opinion roadmap with 45-day DepSec reporting cadence, USMC clean-audit playbook, and 10 AI/ML remediation plays including anomaly detection, graph reconciliation, and agentic CAP drafting.
Live Federal Obligation Execution Tracker
FeaturedAutomated obligation analytics platform with GitHub Actions pipeline pulling DoD data weekly from USASpending.gov API — agency obligations, federal account breakdowns, top contract awards, object-class trends — then chunking and embedding into pgvector for AI-powered Q&A. Supports TAFS period-of-availability analysis, ULO aging buckets, cancellation risk prediction, SF-132/SF-133 comparison, and ADA/FFMIA compliance red-flag detection.
Data Science & Analytics
AI Agent Lifecycle Engineering Framework
FeaturedInteractive 10-stage production AI engineering pipeline — Define Goal → Select Model → Build Knowledge (RAG) → Design Tools (MCP/function schemas) → Agent Loop (ReAct: Thought→Action→Observation) → Add Memory → Orchestrate (multi-agent coordinator/supervisor/A2A) → Evaluate (LLM-as-Judge, golden set) → Guardrails → Deploy & Scale (canary rollout, CI/CD, observability). Built in AXIOM as a hands-on engineering reference grounded in Google 5-Day AI Agents Intensive patterns.
Journal Voucher Anomaly Detection & Graph Reconciliation
FeaturedML pipeline applying isolation forests and autoencoders to surface the 5,665+ unsupported DoD accounting adjustments ($859B, FY2025) for audit triage before quarter close. Graph database with entity resolution for automated buyer/seller intragovernmental transaction matching — flags unmatched pairs for human review. Derived from production AI/ML remediation plays mapped to DoD DODIG-2026-032 material weaknesses 18 and 19.
DoD Spending Semantic ETL & RAG Pipeline
FeaturedEnd-to-end data engineering pipeline: GitHub Actions cron pulls live DoD obligation data from USASpending.gov API (agency totals, federal accounts, budget functions, top contracts) every Sunday → chunked and embedded into Neon pgvector → exposed via RAG-powered AI chat for natural-language obligation analysis. Combines time-series XGBoost cancellation-risk modeling with USSGL TAFS period-of-availability tracking and ADA anti-deficiency risk flagging.
Get In Touch
Have a question or want to work together? Send me a message and I'll get back to you as soon as possible.