Project 01 / 18
AgentBudget Lab
A planning workspace for estimating, comparing, and optimizing the cost, latency, and budget risk of multi-step AI workflows before deployment.
Details
AgentBudget Lab
Highlights
- Deterministic per-step cost, token, retry, cache, latency, and monthly-volume estimation
- Explainable budget-risk scoring, cost-driver findings, and ranked optimization recommendations
- Editable what-if scenarios for model, traffic, retry, cache, and workflow assumptions
- Anonymous localStorage demo and authenticated Supabase workflow persistence
- Editable model-price catalog with provenance and verification warnings
Project 02 / 18
ApplyGraph
An evidence-first application workspace that maps job requirements to candidate proof, explains fit and gaps, and produces source-linked editable materials.
Details
ApplyGraph
Highlights
- Explainable requirement-to-evidence graph with ranked matches and critical-gap detection
- Deterministic fit scoring that measures supplied evidence coverage rather than hiring probability
- Eight editable application outputs with requirement and evidence provenance
- Anonymous browser-local packet and authenticated Supabase workspace architecture
- Owner-scoped RLS, composite graph foreign keys, and transactional packet quota enforcement
Project 03 / 18
AttentionTelemetry
A local-first attention dashboard that turns manual focus check-ins into explainable task-fit, recovery, and daily planning suggestions.
Details
AttentionTelemetry
Highlights
- Manual check-ins and focus sessions with persistent local history
- Deterministic focus, load, distraction, recovery, and task-fit scoring
- Evidence-labeled focus windows and simple recurring patterns
- Recovery-aware daily plans bounded by available time
- Responsive dark/light dashboard with signal inspection and report previews
Project 04 / 18
Axiomizer
An interactive research learning platform that transforms scientific papers into first-principles explanations, visual learning flows, quizzes, and simulations.
Details
Axiomizer
Highlights
- Transforms research papers into interactive slide-based learning experiences
- Focuses on first-principles understanding rather than simple summarization
- Planned support for quizzes, simulations, and visual explanations
- Designed to support future AI coaching and personalized learning
- Built as a long-term research and portfolio platform
Project 05 / 18
Bestpicky
Bestpicky is a deployed office-chair decision demo with ranked guides, evidence records, and transparent product scoring.
Details
Bestpicky
Highlights
- Versioned deterministic scoring with explicit evidence confidence
- Missing dimensions excluded and remaining weights renormalized
- PostgreSQL full-text search with trigram typo fallback
- Server-rendered guides and product pages with SEO metadata
- Retailer-neutral catalog and scoring isolated from affiliate code
Project 06 / 18
BrainArcade
A browser-based collection of arcade, puzzle, trivia, and strategy games built under one unified gaming platform.
Details
BrainArcade
Highlights
- Unified multi-game platform with shared branding and design system
- Interactive Jeopardy implementation using external trivia APIs
- Custom implementations of classic games including 2048, Tetris, Snake, Tic-Tac-Toe, Backgammon, T-Rex Runner, Car Jam, and Chess
- Responsive UI with consistent navigation and reusable styling
- Progressive platform architecture designed for future game additions
Project 07 / 18
ChoiceLoad
A decision-support workspace that turns recurring everyday choices into explainable shortlists, editable defaults, and fallback rules.
Details
ChoiceLoad
Highlights
- Deterministic choice-load and option-fit scoring with explicit explanations
- Constraint-based shortlists that preserve excluded options for review
- Editable default and fallback rules with review dates and recorded applications
- Private browser workspace with templates, reviews, and report snapshots
- Supabase backend code with ownership policies, transactional saves, and quotas
Project 08 / 18
Dataxiom
A data science platform that turns datasets into reproducible studies with documented evidence, model evaluation, and interactive predictions.
Details
Dataxiom
Highlights
- Structured study publishing with a validation checklist
- Dataset provenance and feature dictionaries
- Versioned model registry with SHA-256 artifact verification
- Reproducible heart disease classification study with explicit limitations
- Accessible SVG charts, quizzes, and moderated comments
Project 09 / 18
Dejargonizer
Dejargonizer is a privacy-first web app that translates legal fine print into plain-English risk summaries, highlights, and document Q&A.
Details
Dejargonizer
Highlights
- Built a browser-first legal document review workspace with paste, TXT/MD/PDF ingestion, sample Terms of Service, and clear-session controls.
- Implemented rule-based clause detection for arbitration, class-action waiver, data sharing, content license, auto-renewal, refunds, termination, liability, indemnification, ownership, and deletion/export rights.
- Created high-yield summary tabs for What You Give Up, What You Own, and Red Flags, with risk chips and plain-English explanations.
- Added highlighted-original and side-by-side review modes that map detected clauses back onto the source document.
- Added a private document assistant that answers from detected clauses, with an optional local-model path for private AI assist.
Project 10 / 18
FactoryLogics
FactoryLogics is a browser-native industrial automation lab for building, simulating, commissioning, and troubleshooting PLC-based control systems.
Details
FactoryLogics
Highlights
- Built a browser-native PLC practice lab with editable Ladder Logic, branches, scan execution, live rung states, I/O actuation, and forcing.
- Implemented a guided automation-project workflow spanning hardware, wiring, communication, PLC programming, HMI/SCADA, testing, and fault injection.
- Separated authoritative simulation state into a Web Worker with deterministic integer ticks and inspectable snapshots.
- Added editable LD, Structured Text/SCL, and FBD workspaces with explicit cross-language signal links and a canonical program view.
- Implemented a deterministic package-sorter simulation with HMI commands, alarms, traces, commissioning checks, and boundary-specific faults.
Project 11 / 18
IntentForge
IntentForge turns rough natural-language requests into structured, execution-ready AI prompts with task frameworks, readiness scoring, autosaved history, and optional private cloud accounts.
Details
IntentForge
Highlights
- Feature-based Next.js App Router application with schema-validated prompt generation and refinement APIs
- Provider abstraction supports deterministic mock generation, Gemini, and optional Ollama
- Local-first history saves anonymous prompts to IndexedDB and signed-in prompts to Supabase with RLS
- Email/username password authentication includes signup checks, verification, forgot password, profile controls, and data export
- Production cost controls include per-subject daily limits and a whole-site daily generation cap
Project 12 / 18
KineticNode
KineticNode is a deployed browser-based digital twin prototype for exploring fleet orchestration, induction flow, safety events, and faults on a synthetic fulfillment floor.
Details
KineticNode
Highlights
- Models 840 synthetic drive agents on a 136 by 70 fiducial fulfillment floor.
- Generates 336 destination chutes, 40 induction stations with U-shaped queues, 40 chargers, and inter-chute parking cells.
- Implements browser-local fixed-tick drive, package, battery, routing, fault-cascade, and access-path behavior.
- Provides an inspectable control-room UI with asset telemetry, event logs, scenario modes, layer controls, and trackpad zoom and pan.
- Separates the high-frequency browser simulation from a proposed low-frequency Supabase persistence model.
Project 13 / 18
PaperInsight
AI-powered research paper analysis platform that extracts, structures, summarizes, and stores knowledge from uploaded PDFs.
Details
PaperInsight
Highlights
- Asynchronous PDF processing pipeline using Spring Boot, Redis, and Celery
- Structured research paper extraction into normalized database tables
- LLM-based summarization and metadata generation via Ollama
- Schema-driven JSON validation for extraction quality
- Background worker architecture designed for future cloud deployment
Project 14 / 18
PM Node
PM Node is a portable predictive-maintenance edge device that combines vibration, thermal, temperature, and sound sensing with an on-device TFT UI, local web interface, and recording workflow for machine inspection.
Details
PM Node
Highlights
- Built a multi-sensor predictive-maintenance node around ESP32-S3 for vibration, thermal, sound, and temperature monitoring.
- Implemented both TFT-based on-device navigation and a local web interface with separated live and FFT views.
- Integrated IIS3DWB vibration sensing with trend and FFT analysis workflows.
- Integrated MLX90640 thermal imaging with hotspot, min/max, and thermal palette display logic.
- Added DS18B20 probe support as a separate temperature page with numeric readings and trend plotting.
Project 15 / 18
Portfolio
Personal portfolio and knowledge base platform for projects, Monograph, research, and an AI assistant.
Details
Portfolio
Highlights
- Central hub for projects, research, contact, and Monograph
- Uses a structured KB to power project pages and assistant answers
- Includes secure source IDs to avoid exposing raw filesystem paths
Project 16 / 18
ReasonLoom
ReasonLoom turns permitted tutorial transcripts into source-aware educational article drafts for human review and draft-only publishing to Monograph.
Details
ReasonLoom
Highlights
- Built a deterministic transcript pipeline for TXT, Markdown, SRT, and VTT sources with permission confirmation, cleaning, segmentation, and token estimation.
- Designed a provider-neutral article generation pipeline with concept synthesis, structured drafting, validation, warnings, and immutable article versions.
- Implemented an authenticated editorial workspace and REST-style APIs for transcripts, generation jobs, articles, validation, version restore, export, and publishing.
- Modeled source traceability, concept graphs, prerequisites, article relationships, publication attempts, idempotency records, audit logs, and usage events.
- Created a draft-only Monograph integration contract with bearer authentication, HMAC signing, and idempotency headers.
Project 17 / 18
Simcine
Simcine is a movie and TV discovery platform for semantic title similarity, personal watch libraries, ratings, public taste profiles, and TV episode browsing.
Details
Simcine
Highlights
- Built a Django + Next.js movie and TV discovery platform with user accounts, ratings, watchlist, favorites, watch history, and public profiles.
- Designed privacy-controlled public profile pages at /profile/{username} with public/private sections for watchlist, ratings, favorites, reviews, and history.
- Integrated Google OAuth planning through django-allauth with email verification and connected account status for account security.
- Added TV episode browsing architecture with seasons, episodes, episode detail pages, TMDB import flow, and scrollable episode previews.
- Implemented production-readiness cleanup including header simplification, profile route correctness, typecheck cleanup, footer/legal/cookie planning, and deployment architecture.
Project 18 / 18
VisionBoard
VisionBoard is a future-focused social web app where users create AI-generated aspirational posts and organize them into personal vision boards.
Details
VisionBoard
Highlights
- Instagram-like concept focused on future goals instead of past activities
- AI-generated vision posts from user prompts
- Boards and collections for organizing aspirations
- Free-tier-friendly architecture using Vercel and Supabase
- Designed web-first with possible iOS and Android expansion later