I'm Daniel Nguyen

Led development of a real-time Voice AI Agent and full supporting infrastructure for a self-storage conversational AI platform. Owned the voice pipeline end-to-end — from Twilio telephony ingestion to LLM response — built bi-directional FMS integrations, and stood up the AWS cloud infrastructure and CI/CD workflows serving 4,500+ facilities.

• ▹Architected a real-time Voice AI Agent for inbound tenant support using Python, FastAPI, and WebSockets, ingesting Twilio Media Streams (base64 mulaw) with Whisper (STT) for speech recognition, enabling seamless barge-in with sub-500ms end-to-end latency and reducing live-agent routing by 40% QoQ.
• ▹Implemented Redis Semantic Caching for high-frequency tenant intents, bypassing OpenAI API inference for 35% of inbound calls and cutting p95 latency by 600ms.
• ▹Designed NLP intent classification and routing logic using LLM function calling/tooling that maps caller utterances to autonomous actions, escalation paths, or knowledge-base lookups based on confidence thresholds and tenant context.
• ▹Built an Embeddings-powered knowledge retrieval system using pgvector and RAG (Retrieval-Augmented Generation) to ground agent responses in facility-specific FAQs, policies, and operational data.
• ▹Implemented a PII Redaction middleware pipeline and scheduled asynchronous post-call jobs via Celery/Redis broker for transcript generation, call scoring, and CRM updates — ensuring PCI compliance without blocking the real-time voice pipeline.
• ▹Engineered a bi-directional FMS sync engine integrating StorEdge and SiteLink REST APIs with distributed locking and conflict resolution, ensuring idempotent writes across concurrent updates for 50,000+ storage units.
• ▹Integrated real-time inventory lookup and access code retrieval from FMS systems and developed a secure Action Layer with Stripe Connect for autonomous split-payment processing via OAuth2/JWT-secured API calls, documented with OpenAPI/Swagger specs.
• ▹Built omni-channel workflows dispatching SMS/email follow-ups via Pub/Sub event triggers and designed seamless escalation from Voice AI to live agents with full conversation context summaries, reducing agent ramp-up time on escalated calls.
• ▹Built a Next.js Live Monitor and property-manager dashboard (React, Tailwind CSS, React Query) using Server-Sent Events (SSE) for real-time call transcripts, with LLM-powered sentiment scoring to flag negative tonal shifts and configurable toggles for AI behavior controls.
• ▹Configured CloudWatch dashboards, custom metrics, and alerting for voice pipeline health (WebSocket latency, LLM response time, FMS sync failures), enabling proactive incident response with distributed tracing.
• ▹Engineered a telephony emulation suite to mock Twilio Media Streams and FMS state, with Pytest-based integration and load/latency tests validating sub-500ms p95 under simulated peak traffic for full end-to-end regression testing.
• ▹Provisioned AWS infrastructure (ECS, RDS, ElastiCache, S3, CloudWatch, IAM, VPC) using Terraform IaC, enabling reproducible deployments across staging and production environments.
• ▹Built CI/CD pipelines with GitHub Actions for automated linting, testing, and Docker image builds with zero-downtime deployments to AWS ECS; orchestrated local development with Docker Compose to reduce environment drift.
• ▹Established feature flags and canary rollout strategies for safe incremental releases; managed Linux deployment targets and wrote Bash automation scripts for log rotation, health checks, and database migrations.

Tech Stack: Python, FastAPI, WebSockets, Next.js, React, TypeScript, Tailwind CSS, React Query, Redis, PostgreSQL, pgvector, Celery, OpenAI API, Twilio Voice/Media Streams, Whisper (STT), Stripe Connect, SSE, OAuth2/JWT, OpenAPI/Swagger, Docker, Docker Compose, AWS (ECS, RDS, ElastiCache, S3, CloudWatch, IAM, VPC), Terraform, GitHub Actions, Pytest, Feature Flags, Bash, Linux

Built core agent infrastructure for a conversational AI platform serving enterprise brands. Owned the ReAct agent runtime, vector search retrieval layer, multi-model orchestration, and developer tooling powering multi-step customer workflows across chat, voice, and messaging channels.

• ▹Built the core ReAct (Reason + Act) event loop in Python with transactional state management, enabling agents to reliably execute multi-step workflows with automatic retries, rollback, and idempotent action execution.
• ▹Designed multi-tenant PostgreSQL + pgvector schemas for enterprise knowledge retrieval, tuning partial indexes and HNSW vector search to query millions of rows at single-digit ms p95 latency.
• ▹Built a RAG pipeline with custom chunking, embeddings (stored in pgvector), and hybrid search (keyword + vector search) to ground agent responses in brand-specific knowledge bases and policy documents.
• ▹Developed a Tool Registry that converts Python type hints into OpenAI API-compatible function calling schemas at runtime, enforcing strict input/output validation for all backend tool calls.
• ▹Implemented multi-model orchestration logic enabling agents to dynamically route between OpenAI, Anthropic API, and open-source LLMs based on task complexity, latency requirements, and cost constraints.
• ▹Built a React Trace Visualizer with TypeScript to render step-level agent execution graphs (tool calls, retrieval context, reasoning chains, latency) for faster debugging and prompt engineering iteration.
• ▹Engineered a deterministic action execution layer with gRPC service communication, ensuring all system-of-record mutations (CRM updates, order modifications) are auditable and compliant with enterprise security policies.
• ▹Created an automated Evals pipeline using Pytest to run regression tests against a golden dataset of conversation scenarios, catching prompt regressions and hallucinations before every deployment.
• ▹Integrated PII redaction middleware to scrub sensitive customer data (SSNs, credit cards) from prompts before sending to external LLM providers, ensuring data governance compliance.
• ▹Provisioned AWS infrastructure (ECS, RDS, Lambda, S3, VPC) using Terraform IaC and Kubernetes orchestration, implementing ephemeral preview environments reducing staging spin-up from days to minutes.
• ▹Implemented feature flags and canary rollout strategies with monitoring/alerting to safely deploy experimental agent behaviors to specific enterprise cohorts without production downtime.
• ▹Built conversation analytics and observability tooling with distributed tracing to track agent metrics (resolution rate, CSAT, hallucination rate, latency) with real-time dashboards.
• ▹Developed prompt engineering workflows with few-shot examples, in-context learning, and self-consistency techniques across brand-specific agent configurations, systematically improving response accuracy and reducing hallucination rates in production deployments.

Tech Stack: Python, React, TypeScript, PostgreSQL, pgvector, OpenAI API, Anthropic API, RAG, Embeddings, gRPC, AWS (ECS, RDS, Lambda, S3, CloudWatch, VPC), Terraform, Docker, Kubernetes, Redis, Pytest, GitHub Actions, Feature Flags

Worked across the full stack of a fintech SaaS platform automating Accounts Receivable and Treasury processes. Decomposed monolithic systems into event-driven microservices, built high-throughput payment data pipelines, and delivered front-end tools for AR analysts.

• ▹Decomposed a monolithic payment processor into event-driven microservices using Apache Kafka with Pub/Sub topic partitioning, enabling asynchronous high-volume transaction processing with guaranteed delivery and fault isolation.
• ▹Developed a high-concurrency Java Spring Boot service to parse and normalize unstructured payment data from PDFs, EDI files, and bank remittances at a throughput of 10,000+ documents/hour, using background jobs for async processing.
• ▹Engineered a custom React virtualized grid with Redux Toolkit to render 100,000+ rows of financial data with zero performance degradation, replacing a legacy JSP interface and improving analyst productivity.
• ▹Optimized multi-terabyte SQL queries by implementing table partitioning, covering indexes, and query plan analysis across the PostgreSQL data warehouse, reducing report generation from minutes to sub-seconds.
• ▹Implemented distributed locking via Redis to ensure data integrity and prevent race conditions when multiple analysts modify the same invoice record simultaneously, following idempotency patterns.
• ▹Collaborated with Data Science to deploy a Propensity-to-Pay predictive model, building the low-latency REST inference API with OpenAPI/Swagger docs that served real-time credit risk scores to the collections workflow.
• ▹Built automated cash application matching algorithms that reconcile incoming payments against open invoices using fuzzy matching on remittance data, reducing manual intervention.
• ▹Engineered ETL pipelines to ingest and normalize payment data from 50+ banking formats (BAI2, MT940, CAMT.053), standardizing heterogeneous financial data into a unified SQL schema.
• ▹Built SPA dashboard components in React for AR analysts to monitor Days Sales Outstanding (DSO), aging buckets, and collection effectiveness using GraphQL queries for flexible data fetching.
• ▹Led the migration to a global Redux store with normalized entity state, improving data consistency and reducing redundant API calls across the financial application suite.
• ▹Wrote comprehensive integration tests using JUnit and Mockito, increasing code coverage to 90%+ across critical payment processing paths and ensuring SOC 2 compliance.
• ▹Implemented role-based access control (OAuth2/JWT) and audit logging across the platform to meet enterprise security requirements and support SOC 2 Type II certification.
• ▹Designed event-sourced audit trails for all payment mutations, enabling full transaction replay and reconciliation reporting required for SOC 2 Type II compliance and annual financial audits across 600+ enterprise clients.

Tech Stack: Java, Spring Boot, Apache Kafka, React, Redux Toolkit, JavaScript, TypeScript, PostgreSQL, Redis, MongoDB, GraphQL, REST APIs, OpenAPI/Swagger, Python, JUnit, Mockito, Jenkins, AWS, Docker

Developed front-end frameworks and backend microservices for BMC's enterprise ITSM platform (Remedy/Helix). Migrated legacy services to RESTful architectures and established the team's first automated testing and CI/CD practices.

• ▹Developed a custom Angular directive framework to dynamically render complex ITSM forms based on JSON schemas, enabling no-code workflow building for IT administrators and reducing form development time by 70%.
• ▹Migrated legacy SOAP/XML endpoints to RESTful Node.js (Express.js) microservices, implementing circuit breakers (Hystrix patterns) to prevent cascading failures and improve system resilience under load.
• ▹Built a real-time log aggregation pipeline using MongoDB for server log ingestion, enabling monitoring of automation success rates, incident resolution times, and service health metrics.
• ▹Automated build and deployment pipelines using Jenkins CI/CD, reducing manual release times by 30% and enabling one-click staging deployments with rollback capabilities on Linux servers.
• ▹Built a reusable Angular component library (with HTML/CSS and JavaScript) for ITSM form elements (approval chains, SLA indicators, escalation triggers), standardizing UI patterns across the Remedy suite.
• ▹Developed RESTful API endpoints in Node.js (Express.js) for ITSM workflow automation, enabling third-party integrations with monitoring tools, ticketing systems, and CMDB data sources.
• ▹Developed server-side middleware for request validation, authentication, and rate limiting across the Node.js microservices layer, improving API security posture and performance optimization.
• ▹Implemented ITIL-aligned change management workflows with automated approval routing, conflict detection, and release scheduling, reducing change-related incidents across managed infrastructure.
• ▹Mentored junior developers on asynchronous JavaScript patterns (Promises, async/await, event loops), conducting code reviews that improved overall team velocity by 15%.
• ▹Implemented unit testing frameworks using Jasmine and Karma, establishing the team's first continuous testing protocol for the frontend codebase with Git-hook-based pre-commit validation.
• ▹Developed RESTful webhook integrations between Remedy and third-party monitoring tools (Nagios, SNMP traps), enabling automated incident creation and escalation from infrastructure alerts with deduplication logic.
• ▹Created interactive Angular dashboards with real-time SLA compliance tracking, surfacing MTTR, first-call resolution, and backlog metrics from MongoDB aggregation queries for operations leadership.

Tech Stack: JavaScript, TypeScript, Angular, Node.js, Express.js, MongoDB, REST APIs, SOAP/XML, HTML/CSS, Jenkins, Jasmine, Karma, Hystrix, JSON Schema, Linux, Git, Bash