00 / OVERVIEW
FusionLex AI Platform
01
TECHNICAL PORTFOLIO
Architectures I've built.
Selected AI systems and software architectures designed and implemented across DeepTech, conversational intelligence, agentic AI, machine learning and cloud-native systems.
01 / DEEPTECH
FusionLex
Conversational & Behavioral Intelligence Platform
FUSIONLEX
−
+
FusionLex
Conversational & Behavioral Intelligence Platform
CONTEXT
DeepTech platform designed to transform spoken conversations into structured behavioral, conversational and business intelligence. The platform combines speech processing, NLP, behavioral modeling and large language models to extract actionable intelligence from real-world conversations.
ROLE
Founder & CTO — responsible for AI architecture, software architecture, infrastructure, model deployment, experimentation and productization.
GLOBAL ARCHITECTURE
High-level view of the FusionLex production AI platform.
TECHNICAL VIEWS
Selected system perspectives
01
Cloud Architecture
AWS infrastructure supporting asynchronous AI workloads, storage and scalable inference.
02
LLM Inference
Custom Rust-based TGI serving Llama 3.3 70B with optimized batching, GPU inference and production model serving.
03
Near-Hexagonal Architecture
Speech-to-text, diarization, speech analytics and conversational processing.
04
Semantic Retrieval
Embeddings, vector search, semantic retrieval and grounded generation.
05
Production Infrastructure
Containers, queues, GPU workloads, asynchronous processing and cloud-native services.
06
AI Observability
Model evaluation, tracing, monitoring and operational metrics across the AI pipeline.
CONVERSATION INTELLIGENCE
From raw audio to structured conversational signals.
01
Conversation Extraction
Audio ingestion, transcription, speaker diarization and extraction of conversational structure.
02
Behavioral Intelligence
Transformation of conversational signals into behavioral, emotional and interaction-level representations.
Python · Rust · FastAPI · Angular ·
AWS · Docker · Redis · Kubernetes ·
Whisper · Llama · MLflow · LangSmith ·
Milvus · OpenTelemetry
02 / FUSIONLEX · AGENTIC AI
Market Intelligence
Conversation Intelligence → Market Intelligence
INTELLIGENCE
−
+
Market Intelligence
Conversation Intelligence → Market Intelligence
CONTEXT
An extension of the FusionLex platform designed to transform conversational data into broader market intelligence. Information extracted from conversations can be compared against external discussions, market signals and public information.
OBJECTIVE
Extract structured information from conversations, identify emerging topics, retrieve relevant external information and surface competitors, trends, positioning and potential market opportunities.
AGENTIC MARKET INTELLIGENCE
From conversational signals to external market intelligence.
00 / OVERVIEW
Conversation → Market Intelligence
CONVERSATION
Extracting structured intelligence from conversational data.
01
Conversation Extraction
Transcription → topics → entities → business signals.
02
Signal Normalization
Normalize extracted topics, entities, claims and business signals into structured representations.
MARKET INTELLIGENCE
Expanding internal conversational signals into external market knowledge.
01
Near-Hexagonal Architecture (External Retrieval)
Semantic retrieval across public discussions, market information and external signals.
02
Competitive Intelligence
Competitor discovery, positioning, topics and market signals.
AGENTIC RESEARCH
Planning, retrieval, synthesis and evidence-backed generation.
01
Agentic Research
Planning → retrieval → synthesis → evidence-backed report generation.
02
Quality Evaluation
Grounding, citation coverage, claim verification and unsupported-claim detection.
Python · LangGraph · Amazon Bedrock ·
Titan Text Embeddings V2 · S3 Vectors ·
RAG · Agentic AI · Semantic Search ·
LLM Evaluation
03 / INTELLIGENT AUTOMATION
Reflex
Intelligent Client Request Triage
REFLEX
−
+
Reflex
Intelligent Client Request Triage
CONTEXT
Intelligent system designed to process incoming requests and questions from potential clients. The system combines supervised machine learning and agentic reasoning to understand the intent of each request and determine the appropriate next action.
OBJECTIVE
Automatically classify incoming requests, identify missing information, generate clarification questions, qualify the request and route it to the appropriate operational workflow.
ML + AGENT ARCHITECTURE
Intelligent qualification pipeline for potential client requests.
00 / OVERVIEW
Reflex Intelligent Triage
REQUEST UNDERSTANDING
Understanding and qualifying incoming client requests.
01
Intent Classification
Transformer-based classification of incoming client requests.
02
Clarification
Detection of missing information and generation of clarification questions.
AGENTIC WORKFLOW
From classification to qualification and operational routing.
01
Request Qualification
Context validation, business rules and request qualification.
02
Agentic Workflow
Classification → clarification → qualification → routing.
PRODUCTION
Cloud deployment of the intelligent triage system.
01
Cloud Deployment
API Gateway · Lambda · SageMaker · FastAPI.
02
Operational Routing
Mapping qualified requests to queues, priorities and downstream operational workflows.
Python · DistilBERT · Transformers · Pandas · Matplotlib · TensorFLow · Keras ·
SageMaker · Lambda · API Gateway ·
AWS