I build security-conscious software for education, operations, and AI-assisted workflows. My work is strongest where a product needs explicit data boundaries, reliable automation, and honest operational status.
$ cat capabilities.md
# where I create the most leverage
Secure Backend Systems
SYSTEMSI turn messy product requirements into explicit APIs, data boundaries, and services that are easier to verify and operate.
- >Rust and Go services with typed contracts and clear ownership boundaries
- >Tenant isolation, authentication, authorization, audit trails, and negative-path testing
- >Architecture decisions documented with current limits instead of inflated claims
Operational Workflow Automation
OPERATIONSI replace fragile manual coordination with software that makes state, permissions, and recovery paths visible.
- >Education, scheduling, reporting, inventory, and internal operations workflows
- >Idempotent jobs, preview-first execution, integrations, and human recovery paths
- >Product decisions shaped by the people who actually run the process
AI-Assisted Execution
AI / AUTOMATIONI use AI where it improves a workflow, then add structure, validation, and repair around probabilistic output.
- >Structured generation instead of unbounded text-only prompts
- >Document ingestion, retrieval, repair loops, and human-in-the-loop review
- >A bias toward observable, replaceable automation rather than magic demos
$ cat proof-points.json
# the evidence behind the positioning
$ tree ~/projects --dirsfirst
# current systems, shipped products, and selected work
Current Systems
The projects that best represent my current backend, security, and architecture direction.
Vendryx Commerce
[SYSTEM]A Rust/PostgreSQL multi-tenant commerce platform with merchant administration, storefronts, and a partially completed operator control plane.
- problem:
- How can merchants share an application and database without relying solely on application-level tenant filters?
- architecture:
- Modular Rust services, PostgreSQL FORCE ROW LEVEL SECURITY, transaction-scoped tenant context, and separate storefront/admin frontends.
- proof:
- Source migrations implement fail-closed tenant policies; integration and release gates exist, but the latest available CI runs failed.
- -> FORCE RLS and tenant-context checks are implemented in SQL migrations, not just specified in documents
- -> Commerce phases are marked complete internally, but the transitional control plane is unfinished
- -> No verified live merchant adoption or successful current CI run is claimed
Hall Security
[SYSTEM]A Rust-based research instrument for hall-entry identity verification and security monitoring, deliberately constrained to observe-only behavior.
- problem:
- How can a security-monitoring prototype collect identity evidence without being trusted to actuate a physical access control?
- architecture:
- Rust services for vision, control, dashboard, and shared types; PostgreSQL, mTLS, hash-chained audit, and a disk-backed outbox.
- proof:
- Rust event-delivery code, observe-only invariants, and security gates exist. Latest CI and GPU release qualification remain unresolved.
- -> Separates observation from actuation: it cannot open doors or grant access
- -> Uses TLS/mTLS, shared Rust types, audit records, and a disk-backed outbox
- -> Includes invariant tests, generated TypeScript types, and documented release blockers
- -> Research prototype: thresholds remain uncalibrated and a qualified release is outstanding
LogBook Institution
[SYSTEM]A Go and Next.js academic LMS/ERP foundation exploring Moodle-scale institutional workflows with explicit tenant boundaries and a modular backend.
- problem:
- How can a large academic workflow be decomposed into a maintainable backend without weakening tenant boundaries?
- architecture:
- Go with chi and pgx, PostgreSQL migrations, JWT-derived organization context, and a Next.js application surface.
- proof:
- JWT rejection tests, modular Go APIs, and SQL migrations exist; the latest visible CI run failed and the teaching core remains incomplete.
- -> JWT middleware tests cover invalid claims, expiry, signatures, and authorization states
- -> Derives org_id from verified JWT claims rather than trusting request bodies
- -> Uses tenant isolation tests and negative-capability tests for security-sensitive paths
- -> An August 2026 project audit says the teaching core cannot yet run a university semester
LogBook Professor Workspace
[SYSTEM]A multi-tenant classroom-management SaaS and educator workspace built around RLS, offline-first synchronization, and operational workflows.
- problem:
- How should educator workflows behave when organization boundaries and low-connectivity synchronization are both non-negotiable?
- architecture:
- Next.js and React with Supabase/PostgreSQL, org_id-based RLS, PowerSync, and property, security, and load tests.
- proof:
- A private repository with explicit tenant rules, offline-first direction, and tests around security and synchronization behavior.
- -> Models academic work across organizations with org_id-based data boundaries
- -> Combines RLS, offline-first sync, and property/security/load testing
- -> The repository is private; the public story focuses on architecture and trade-offs
Products & Tools
Tools and products; maturity and current limitations differ by project.
IPE-24 Classroom
[PRODUCT]A class operating system that centralizes announcements, schedules, resources, and AI-supported student workflows for an active academic community.
- problem:
- How can one student batch replace fragmented chat threads, drive links, routine changes, and repeated questions with a dependable source of truth?
- architecture:
- Next.js, PostgreSQL, Prisma, Redis, Telegram automation, and FastAPI-backed workflow support.
- proof:
- Built for a university cohort; the reported 60+ student users are not independently verified analytics.
- -> A major product that deepened my software-building focus in June 2025
- -> Reported use by 60+ students; independently audited usage metrics are unavailable
- -> Shows end-to-end ownership: problem discovery, product decisions, deployment, and iteration
Email Automator
[PRODUCT]A Gmail-based bulk email tool for personalized announcements, grades, and reports with preview-first safety and structured data merging.
- problem:
- How can an operator send sensitive personalized messages at scale without trusting a blind batch or a third-party mail database?
- architecture:
- Next.js and NextAuth around Google OAuth, Gmail API delivery, CSV normalization, and Handlebars templates.
- proof:
- Public source implements OAuth-based Gmail sending and preview. Its import endpoint has unresolved authorization and retention concerns.
- -> Treats high-stakes sending as a reviewable workflow rather than a blind batch job
- -> Merges messy CSV data into deterministic recipient workflows
- -> Uses delegated Gmail sending; the import-data path remains a documented security limitation
SlideCrafter AI
[PRODUCT]A local-first AI presentation compiler that turns raw context into structured, editable decks with validation and repair around model output.
- problem:
- How can probabilistic model output become a valid, editable PowerPoint without trusting the model to control every layout detail?
- architecture:
- React, TypeScript, and Vite around a structured intermediate representation and deterministic PptxGenJS compilation.
- proof:
- A public compiler workflow with validation and bounded repair before deck export.
- -> Frames deck generation as a compiler pipeline instead of prompt-only text generation
- -> Validates layout structure before export and supports bounded repair loops
- -> Keeps the final compilation step close to the user for fast iteration
Selected Products
Earlier builds that show product judgment, domain range, and the path toward deeper systems work.
The Matrix Hours
[SELECTED]A watch storefront shaped around curated inventory, owner-friendly operations, and a low-friction WhatsApp buying flow.
- problem:
- How can a small high-trust seller publish curated inventory without inheriting heavyweight commerce operations too early?
- architecture:
- Next.js and React with Google Sheets as the operator-friendly inventory source and WhatsApp as the buying handoff.
- proof:
- A shipped storefront aligned to the seller's current process rather than a generic ecommerce template.
- -> Uses a normalization layer around Google Sheets instead of forcing heavyweight commerce infrastructure
- -> Balances trust-building presentation with operational simplicity
Tutors Hub
[SELECTED]A tutoring marketplace designed as a workflow system, with matching, verification, routing, and human recovery paths.
- problem:
- How can a tutoring request move through changing requirements, verification, matching, and payout without losing operational context?
- architecture:
- Next.js for product surfaces, Spring Boot for workflow ownership, PostgreSQL/Supabase for relational state, and Firebase notifications.
- proof:
- Workflow specifications and private implementation exist; the currently deployed feature set has not been independently validated.
- -> Models matching and verification as business state transitions
- -> Separates onboarding UX from security-sensitive backend workflows
- -> Makes operational recovery part of the product rather than an afterthought
Krishi Dristi
[SELECTED]An offline-first agricultural intelligence prototype exploring crop issue detection under fragmented-farm and low-connectivity constraints.
- problem:
- How can crop intelligence work for fragmented plots when connectivity, hardware, and field assumptions are all constrained?
- architecture:
- Python, OpenCV, TensorFlow, and MobileNetV2 experiments designed around local processing and later synchronization.
- proof:
- A constraint-driven prototype that documents what remains unvalidated instead of presenting a notebook as a deployed service.
- -> Starts from field constraints instead of assuming reliable connectivity or uniform plots
- -> Combines classical vision and ML experiments where each is useful
- -> Clearly remains a prototype rather than a production agricultural product
$ cat skills.json | jq '.evidence'
# tools connected to work I can point to
Backend & Data
Security & Reliability
Infrastructure & Integration
Applied Automation
$ find './case-studies' -name '*.md' | sort
# technical evidence, trade-offs, and current limits
Read the deeper write-ups for product context, architecture, security boundaries, difficult decisions, verification, and the work that remains. The goal is to make the engineering evidence inspectable—not to make every project sound finished.
$ cd ./case-studies && ls -la _$ cat contact.sh
# open to backend systems, automation, and difficult product problems
# If you need someone who can move from ambiguous workflow to a tested, explainable system, reach out.