
VoiceDesk is a real-time AI-powered customer support agent that transforms how enterprises handle customer interactions. Instead of typing, customers simply speak — and VoiceDesk listens, understands, and responds intelligently in seconds. HOW IT WORKS The user clicks the microphone button in the web interface. Audio is captured at 16kHz directly in the browser and streamed via WebSocket to a FastAPI backend. The backend pipes the audio stream to the Speechmatics Real-Time API, which returns partial transcripts as the user speaks (visible live in the UI) and final transcripts when the utterance is complete. The final transcript is then sent to Claude Sonnet, which acts as an expert customer support agent and generates a concise, actionable response. The entire exchange — transcription + AI response — appears in a clean conversation log with latency metrics. KEY FEATURES - Real-time partial and final transcription powered by Speechmatics RT WebSocket API - Live audio waveform visualizer for immediate feedback - Claude Sonnet AI responses optimized for enterprise customer support tone - Session analytics: exchange count, average response latency, word count, session duration - Single-page web application — no installation required for end users - Production deployment on Vultr cloud infrastructure USE CASES VoiceDesk targets enterprise customer support teams, call centers, and any business that wants to deflect repetitive support queries with an autonomous voice agent. TECHNICAL STACK - Speechmatics Real-Time API (WebSocket, PCM float32 audio, 16kHz) - Anthropic Claude Sonnet for response generation - Vanilla JS frontend with Web Audio API for waveform rendering - Deployed on Vultr Ubuntu VM WHAT MAKES IT DIFFERENT Most voice AI demos are disconnected pipelines. VoiceDesk is a single cohesive system where speech, transcription, and AI reasoning happen in one continuous real-time flow. The result feels instant — because it is.
19 May 2026

TEK TEK Agent is an AI-powered web platform built with FastAPI that unifies 5 specialized agents designed to help developers and engineering teams fully understand, audit, and improve their codebases — all powered by the IBM Bob large language model. Bob Guard automatically reviews GitHub pull requests through 5 analytical lenses (regressions, missing tests, security vulnerabilities, pattern conformity, and documentation), delivering a structured risk-scored review directly to your PR. Bob Archaeologist reverse-engineers legacy codebases by detecting architectural patterns, business entities, dangerous code zones ("dragons"), reconstructed architecture decision records (ADRs), and generates a custom onboarding path for new developers. TechDebt Radar quantifies your technical debt in real USD using static analysis tools (Radon, Vulture, pip-audit), scanning 7 dimensions including cyclomatic complexity, dead code, outdated dependencies, and CVE-flagged packages — then produces a prioritized refactoring sprint plan with payback timelines. Bob Standup automatically generates a full 4×3 matrix of reports from your GitHub activity — 4 audiences (developers, managers, clients, retrospectives) across 3 languages (English, French, and Haitian Creole) — with optional distribution via Slack or email. CodeStory transforms any codebase into animated SVG stories with voice narration, tracing feature execution flows across architectural layers in 3 tones (technical, casual, kid-friendly) and 3 languages, making codebases accessible to any audience. All agents share a unified async job pipeline with real-time WebSocket progress tracking, disk-cached IBM Bob API calls, SQLite persistence, and a clean dashboard aggregating metrics across every analysis run.
17 May 2026

MarketMind is an AI-powered crypto market intelligence platform built on a multi-agent Gemini pipeline. Instead of relying on a single model call, MarketMind orchestrates four specialized AI agents that work sequentially to deliver institutional-grade market analysis in real time. How It Works Agent 1 — Data Collector aggregates live market data from CoinGecko (prices, volumes, market caps, 7-day OHLC history for BTC, ETH, BNB, SOL, XRP) and fetches the latest crypto news from multiple RSS feeds and web sources via a built-in scraper engine. Agent 2 — Pattern Analyst (Gemini) performs deep technical analysis: it identifies market structure (bull/bear/range), detects anomalies in volume and price action, spots support/resistance levels, and assigns a global risk level (FAIBLE / MODÉRÉ / ÉLEVÉ / EXTRÊME). Agent 3 — Sentiment Analyst (Gemini) processes up to 20 news articles using NLP to extract: overall market sentiment (BULLISH/BEARISH/NEUTRAL), per-asset sentiment scores, emerging narratives, macro risks, and a Fear & Greed index. Agent 4 — Intelligence Orchestrator (Gemini) synthesizes both analyses into a full Market Intelligence Brief with probabilistic forecasting (bull/bear/neutral %) for 24h and 7-day horizons, plus actionable trade signals with 3 Take Profit and 3 Stop Loss levels per signal. Key Features MarketMind demonstrates that a small, well-orchestrated multi-agent system can produce richer, more reliable intelligence than a single large prompt — by separating concerns across specialized agents that each excel at their specific task.
19 May 2026