TD Monogram
Tamas Deak

Search & Filter Timeline

Showing 16 of 16 projects
  • 2026

    JobAgent — Autonomous Career CRM & Statutory Compliance Engine

    pythonfullstackaitoollocal-first

    A local-first, autonomous career dashboard that turns chaotic job hunting into a single intelligent cockpit: 1-click posting archive with Markdown + PDF snapshots, multi-protocol email triage with genuine interview detection (even in German), and automatic Agentur für Arbeit proof reports. Built on the AWS Strands Agents SDK — zero PII leaves your machine.

    JobAgent is an autonomous background agent for job seekers, submitted to the Agents for Humans hackathon (AWS Strands Agents SDK). It turns the chaos of job hunting — expiring postings, scattered emails, hidden interview invites — into a single intelligent cockpit with zero PII leaving the machine.What it does:
    - Dual-asset archiving — 1-click browser extension captures a clean Markdown brief and a pixel-perfect PDF snapshot before the posting disappears
    - 3-tier email triage — deterministic rules → offline ML classifier → Gemini semantic reasoning detects genuine interview invitations (including German 'Einladung zum Vorstellungsgespräch'), filters recruiter spam/webinars, tracks Applied → Interview → Rejected lifecycles
    - Statutory compliance — automatically compiles the official German Agentur für Arbeit proof table (Eigenbemühungsnachweis, §138 SGB III) as print-ready PDF
    - Agent-to-Agent (A2A) & MCP server — Telegram, Claude Desktop, Cursor and other parent agents query and control JobAgent via standard A2A/SSE endpoints
    - 100% local-first — SQLite CRM + local archive; optional Ollama for true zero-cloud inference    Architecture: Strands Agent Coordinator (Gemini) orchestrating EmailIngestTool → JobArchiveTool → ReportRenderTool over a local SQLite CRM, exposed via FastAPI A2A gateway + MCP and a Chrome Manifest V3 Copilot extension.Quality: 49-test suite (PII/secret scans clean across full git history), adversarial QA pass, DNS-rebinding-guarded loopback gateway.Demo video (2:02, narrated): screencast replayed with Frame Talk's synchronized two-host walkthrough. Code: https://github.com/tdk67/jobagent
    Code▶ Watch demo (2:02)

  • Frame Talk — Multimodal Screen-to-Podcast Studio Engine

    pythonfullstackaimultimodaltool

    Transforms silent developer screencasts and README docs into synchronized two-host technical podcast walkthroughs — Gemini 3.7 Flash vision reads video pixels, Chronos Sync Engine matches dialogue to UI actions with millisecond precision.

    Frame Talk is the Multimodal Screen-to-Podcast Studio Engine, built for the Agentic Cinema Hackathon. It ingests a raw silent .mp4 screencast plus project README.md, and uses Gemini 3.7 Flash native multimodal comprehension to extract granular visual scenes (UI actions, clicks, terminal logs, state transitions) directly from video pixels — no external transcripts.Pipeline: Ingestion & Alignment Agent → Scriptwriter Persona Agent → QA & Pacing Audit Agent → Chronos Sync Engine (TTS + duration meter + freeze calculator) → Compiler Agent (FFmpeg 1080p stitching with permanent holds).Highlights:
    - Scriptwriter persona agents generate organic two-host dialogue between Mark (Lead Systems Architect) and Sarah (Dev Advocate) with natural turn-taking pauses (180-240ms) and zero synthetic timestamps.
    - Chronos Sync Engine synthesizes every line via gemini-3.1-flash-tts-preview at 24kHz PCM, measuring audio duration to the millisecond (duration_ms = pcm_bytes / 48) and computing required_freeze_ms to hold the video timeline at the focal action — audio and screen never desync.
    - Observability via ClickHouse time-series DB + Grafana dashboard tracking turn_index, audio_ms, freeze_ms, and drift.
    - BYOK — no server-side credentials required.
    CodeLive DemoLinkedIn

  • Idea Lint — Adversarial Multi-Agent Venture Validation Engine (FlyRank Capstone)

    pythonaibackendtool

    An impartial AI courtroom for startup ideas — researches with live Google Search grounding, debates across adversarial agents, and outputs an audit-proof, scored PRD. Privacy-first with zero cloud storage and BYOK free-tier support. Google hackathon submission: devpost.com/software/idea-lint

    Idea Lint (formerly VentureBot) is an adversarial, multi-agent venture validation engine built on Google ADK. It takes a raw startup concept and runs it through a panel of specialized agents with deliberately asymmetric information access:
    • Researcher — mines competitor signals and market metrics via live Google Search grounding
    • Creative Ideator — proposes lateral pivot angles
    • Advocate — argues the strongest bull case
    • Critic — red-teams feasibility, moats, and unit economics traps
    • Judge — synthesizes everything into structured 1-10 scores (Novelty, Feasibility, Market Fit) with a PROCEED / PARK / PRUNE verdict
    • PRD Writer + Security Auditor — produce a rigorous, machine-actionable PRD
    V1 was submitted to the Google Gemini Hackathon on Devpost: https://devpost.com/software/idea-lint. For the FlyRank Backend Track Capstone it was re-architected into a production-grade, zero-retention backend: decoupled asynchronous job engine, hardened REST gateway with rate limiting, adversarial agent guardrails, a 161-test deterministic test suite, and browser-local IndexedDB persistence (BYOK). Live demo: https://idea-lint.taskmind-ai.com
    CodeLive DemoLinkedIn

  • BE-08 — PDF Report Generator & Immediate Streaming API

    pythonbackendtool

    Executive analytics & branded PDF reporting engine with single-flight concurrency, idempotency, and an immediate 0ms-TTFB streaming endpoint — deployed on a VPS behind Nginx.

    An executive analytics and PDF reporting engine for the FlyRank Backend Track. Queries the BE-06-Scraper SQLite database (1,000 books), generates FL-05 Identity Kit-branded PDF reports (Inter & JetBrains Mono, Sky Blue #0284c7), served via two modes:
    • Standard endpoint (POST /reports) — returns a file link
    • Immediate streaming endpoint (POST /reports/stream) — streams PDF bytes in 8KB chunks straight from memory/disk with 0ms TTFB on idempotent hits
    Architecture highlights:
    - Clean 4-layer separation (routers → services → repository → SQLite, zero SQL outside the repository layer)
    - Single-Flight asyncio.Lock with double-check pattern preventing races under concurrent generation
    - Same-day idempotency — duplicate requests reuse the existing PDF
    - SQL aggregations computed directly in SQLite with performance indexes (idx_books_category/rating/price)
    - Configurable catalog cap (max_catalog_limit) with truncation badge
    - Jinja2 template decoupling — design changes require zero backend edits
    - Playwright headless Chromium rendering to pixel-perfect A4 bytes, isolated in a thread-pool executor    Includes a live streaming monitor dashboard (GET /demo) with real-time ReadableStream reading, live TTFB and chunk counters. 23 passing pytest tests across 6 modules including a 5-request concurrency test. Deployed on a VPS behind Nginx + Let's Encrypt SSL — live demo: https://taskmind-ai.com/flyrank-be-08/demo
    CodeLive Demo

  • BE-07 — Book Translation API with LLM

    pythonbackendaisecurity

    Production-grade schema-validated translation API with 5-layer prompt injection defense, cost logging, kill switch, and Swagger UI — deployed on a VPS behind Nginx.

    A production-grade, schema-validated API that accepts a book ID and target language (de/fr/it/en), fetches a scraped book record, and uses an LLM (via OpenRouter) to generate clean, structured JSON translations of the book's title and description.Built with FastAPI + the official OpenAI client library — provider-agnostic (local Ollama or cloud OpenRouter).Features:
    - Strict Pydantic validation (extra=forbid, min_length=1)
    - Response match verification + 1 repair retry on schema failure with quarantine logging
    - 30s cumulative wall-clock timeout (HTTP 504) and exponential backoff on rate limits
    - Structured cost logging to JSONL
    - Fail-closed Kill Switch (LLM_ENABLED=false → HTTP 503)    Security: 5-Layer Prompt Injection Defense (OWASP LLM01) — role separation, JSON encoding, anti-hijacking instructions, hard schema firewall, and few-shot security examples — preventing scraped text from hijacking prompt instructions.Deployed on a VPS behind Nginx reverse proxy with Let's Encrypt SSL. 15 passing pytest tests, a 9-case eval suite (100% pass rate), and interactive Swagger UI docs.
    CodeLive Demo

  • 3D Quaternion Calculator

    pythonfrontendtool

    Interactive 3D wireframe cube calculator powered by quaternion rotation — with an exact rational-number calculator side-by-side.

    A dual-mode interactive calculator built with Python and Streamlit.
    • Rational Mode — exact arithmetic using Python's fractions.Fraction class; no floating-point errors (1/3 + 1/6 = 1/2, not 0.4999...)
    • Quaternion Mode — live 3D wireframe cube with RGB coordinate axes, rotated via a calculator-style numpad with real-time perspective projection; scientific degree input for precise pitch / yaw / roll
    Originally created as an AI pair-programming experiment to test how well an AI assistant could design an interactive calculator with exact fraction arithmetic and 4D quaternion math — the 3D viewport demonstrates quaternion operations visually, no game engine required.
    CodeLive DemoLinkedIn

  • BE-06-Scraper — Progressive Multi-Target Web Scraper

    pythonfullstacktool

    Production-grade 3-stage ETL web scraper with politeness ethics layer and Streamlit live demo, deployed on Streamlit Community Cloud.

    A production-grade, extensible web scraping service implementing a 5-stage ETL pipeline (Fetch → Parse → Extract → Clean → Structure → Store) with a strict Politeness & Ethics Layer.Strategy Pattern engine with 3 progressive targets:
    - Static HTML — books.toscrape.com via BeautifulSoup4
    - JSON-LD microdata — dasoertliche.de German B2B leads with OpenStreetMap geocoding
    - Dynamic SPAs — kaggle.com via Playwright headless browser    Includes token-bucket rate limiting, robots.txt compliance, exponential backoff via tenacity, Pydantic v2 validation, and SQLite persistence. The interactive Streamlit dashboard provides live scraping controls, database exploration, record inspection, and audit log viewing. Deployed on Streamlit Community Cloud.
    CodeLive DemoLinkedIn

  • DeutschLinguist: Philology Academy

    typescriptfullstackaieducationgerman

    Interactive educational web app for advanced German literature, linguistics, and stylistic analysis across a 6-month, 24-work curriculum.

    An interactive educational platform for deep German literary study covering a structured 6-month curriculum through 24 canonical works — from Goethe's Faust I to Jenny Erpenbeck's Gehen, ging, gegangen — organized into six thematic blocks: Classical Foundations, Romanticism, Realism, Modernism, Post-War Reckoning, Contemporary Voices.Student-first design:
    - BYOK multi-provider AI tutoring
    - Text-to-Speech audio playback
    - Client-side IndexedDB persistence
    - Markdown AI tutor chat with German philological guardrails    Generates automated philological lectures tailored to each work.
    CodeLive DemoLinkedIn

  • Experimenting with Pi Coding Agent

    frontend

    Maintained this portfolio site using Pi — created a 10-slide interactive presentation showcasing Pi's capabilities versus Claude Code.

    Used Pi as the coding agent to maintain and extend this very portfolio site, including a comprehensive 10-slide HTML presentation exploring Pi's architecture:
    • Minimal core
    • TypeScript extensions
    • 30+ providers
    • Session tree branching
    • Skills & prompt templates
    • Pi's deliberate philosophy of staying out of your way
    Targets technical folks familiar with Claude Code who haven't experienced Pi's extensibility model yet. Pi handled the full workflow — reading documentation, generating the presentation HTML/CSS/JS, integrating it into the Next.js Docker deployment, and updating the portfolio data — all in one continuous session.
    Live Demo

  • Pi A2A Agent Mesh

    pythonbackendinfraai

    Production-grade agent-to-agent communication mesh with dual-model intelligence, 5-layer security, and long-term memory.

    A complete framework for building self-organizing AI agent networks. Each agent runs two pi models in parallel — a cheap one for coordination (gemma-3-12b) and a capable one for real work (deepseek-v4-pro) — with automatic complexity-based switching.Features:
    - 5-layer security hardening (firewall, bearer token, rate limiter, guard rails, audit logging)
    - Cryptographically verified message exchange with HMAC signatures
    - Long-term memory via mem0
    - Relaxed exponential backoff for inter-agent polling
    - Built on the fasta2a Python library implementing the A2A protocol v0.3.0
    - One-command setup script for deploying new agents    Live demo run: 2026-07-27T08:40:24Z. Also see the Pi Coding Agent deep-dive: https://portfolio.taskmind-ai.com/pi-presentation
    CodeLive Demo
Page 1 of 2