Aslan Farboud
TFI Group AI Platform

TFI Group AI Platform

Next.jsTypeScriptMulti-Modal AILLM OrchestrationVector SearchRAGOpenAI GPT-4o/O1Voice AI+4 more

AI-powered platform for tax consultancy featuring multi-modal AI assistants, advanced document processing pipelines, and sophisticated LLM orchestration. Built comprehensive AI workflows including Tina AI (conversational AI with vector search), Dan AI (voice-enabled tax consultant), automated document analysis with parallel processing, and schema-driven AI configuration systems.

This is a proprietary client project. Live demo access is restricted due to confidentiality agreements.

Architected and developed a comprehensive enterprise AI platform for TFI Group, a leading tax consultancy specializing in Housing Development, Land Remediation Relief, R&D tax credits, and Capital Allowances in the UK. The platform demonstrates up-to-date AI application development with multiple sophisticated AI systems working in coordination. Built Tina AI Assistant - an advanced conversational AI with multi-model support, featuring intelligent file referencing using @filename syntax, real-time web search integration, and dual-scope vector stores for personal and organizational knowledge. Implemented streaming responses, conversation persistence, and context-aware personalization. Developed a beta Dan AI Consultation System - a specialized voice-enabled virtual tax consultant with real-time speech-to-text/text-to-speech capabilities, domain-specific persona optimization, and structured consultation workflows mimicking professional tax advisory sessions. Created sophisticated document processing pipelines with multi-stage AI analysis workflows including Python-based PDF extraction, Deepgram audio/video transcription with speaker diarization, intelligent content chunking, parallel AI processing across multiple models, reference validation with fact-checking, and automated report generation with Excel output. Implemented advanced schema management framework for dynamic AI behavior configuration, including customizable question sets, LLM parameter control, writing style rules, version control with rollback capabilities, and training example management for consistent AI outputs. The platform showcases enterprise-grade features including MFA with TOTP-based 2FA, domain-restricted authentication, role-based access control, hierarchical file organization with vector search, cost optimization with token usage tracking, and production-ready error handling with fallback strategies.

Features

  • Tina AI: Multi-model conversational AI with vector search and file referencing
  • Dan AI: Voice-enabled virtual tax consultant with domain expertise
  • Advanced document processing with parallel AI analysis workflows
  • Schema-driven AI configuration with dynamic behavior management
  • Enterprise authentication with MFA and role-based access control
  • Real-time streaming AI responses with cost optimization
  • Sophisticated vector search with dual-scope knowledge bases
  • Multi-stage document analysis with reference validation

Technologies

Next.js 15+TypeScriptOpenAI GPT-4o, O1, O3-miniGemini, Grok, ClaudeVector Stores & RAGDeepgram Speech AIPython Document ProcessingTrigger.dev WorkflowsSupabase (Auth, DB, Storage)Multi-Factor AuthenticationReal-time Streaming APIsCost Optimization SystemsReference Validation AISchema Management Framework