
Daniel's Dev Projects
Software developer
I'm a software developer focused on backend engineering. I work primarily in Python and FastAPI, building async services and REST APIs backed by PostgreSQL and MongoDB, with Pydantic handling validation throughout. Much of my recent work centers on AI systems: RAG pipelines, hybrid dense and sparse retrieval over vector databases such as Qdrant, and agent workflows built with Pydantic AI. I take projects through to production using Docker, Kubernetes, and Google Cloud. On the frontend, SvelteKit is my primary framework, with working experience in Vue. I began my career in network engineering, spending a decade in enterprise and carrier infrastructure before moving into software development.
Real-Time AI Call Auditor
Built a proof-of-concept platform to determine whether LLMs could evaluate live voice calls with low enough latency to provide useful feedback while the conversation was still happening. Supervisors could define dynamic evaluation rubrics for different call scenarios, while agents received real-time feedback showing which required goals had been satisfied and what was still missing.
Streamed live audio chunks over WebSockets for asynchronous transcription and processing
Evaluated conversations continuously against configurable rubrics and call goals
Supported requirements such as introductions, company identification, required legal language, professionalism, and scenario-specific procedures
Built an LLM auditor agent that evaluated transcript updates and detected completed rubric goals
Delivered evaluation results asynchronously to the agent dashboard during the active call
Achieved approximately <2 second end-to-end evaluation latency, demonstrating feasibility for real-time use
Designed a microservices architecture with asynchronous communication and Pub/Sub messaging
Built a supervisor workflow for creating and updating rubrics for different healthcare call scenarios
Created an additional AI customer simulator that could hold a spoken conversation with the agent for testing and training
Streamed transcription, AI responses, evaluation results, and generated voice back to the frontend in real time
Demonstrated potential for both live quality assurance and agent training, giving supervisors immediate control over scripts, requirements, and evaluation criteria
Tech stack: Python · FastAPI · SvelteKit · PydanticAI · MongoDB · Valkey · Pub/Sub · WebSockets · AsyncIO · LLMs · Speech-to-Text · Text-to-Speech · Microservices
Pangea Network Discovery
I designed and built Pangea for Vodafone after existing third-party discovery tools could not reliably handle our large, multi-vendor network. It automated device discovery, data collection, enrichment, and integration across more than 800,000 host IP addresses.
Scanned 800,000+ IP addresses in under 7 hours using Python multiprocessing
Discovered routers, switches, media gateways, LTE modems, and other network devices
Designed a driver-based architecture for communicating with different vendors and device types
Used SSH and SNMP to identify devices, collect metadata, and retrieve configurations
Built parsers and validators to normalize collected data into a consistent network inventory
Enriched discovery data with information from CRM, IPAM, customer, and other internal systems
Developed a REST API for network engineers and internal Vodafone applications
Integrated Pangea with monitoring platforms, data collectors, and service-delivery systems
Automated discovery workflows with minimal human intervention
Tech stack: Python · Flask · Multiprocessing · REST API · SSH · SNMP · Network Automation · Data Parsing & Validation
Hitchat Business
Built a SaaS platform that transformed company documents and internal knowledge into an AI-powered knowledge base, allowing employees to ask questions about policies, manuals, procedures, and other internal content through a conversational interface.
Designed and built end-to-end RAG pipelines for document ingestion and retrieval
Implemented hybrid search combining dense and sparse retrieval
Worked with Pinecone and later Qdrant vector databases
Experimented with multiple document chunking and embedding strategies
Combined retrieval results using Reciprocal Rank Fusion (RRF)
Added cross-encoder reranking to improve relevance before passing context to the LLM
Built backend services for document processing, search, and LLM integration
Deployed and maintained services on Google Cloud
Tech stack: Python · FastAPI · RAG · LLMs · Qdrant · Pinecone · Vector Search · Hybrid Search · Embeddings · RRF · Cross-Encoder Reranking · GCP · Docker · Kubernetes
Cyber Threat Intelligence Platform
Built a cybersecurity threat intelligence platform based on STIX 2, enriching standardized threat data with information collected from third-party APIs and web sources. The data was stored as a graph in ArangoDB to uncover relationships between indicators, threats, infrastructure, and other security entities.
Built backend services and APIs with FastAPI
Ingested and enriched STIX 2 threat intelligence data
Integrated external data sources through APIs and web scraping
Modeled threat relationships using ArangoDB graph database
Deployed services using Docker and Kubernetes
Tech stack: Python · FastAPI · STIX 2 · ArangoDB · REST APIs · Web Scraping · Docker · Kubernetes


