Agentic AI & Orchestration
Multi-step agents with tool calling, memory and deterministic fallbacks for regulated domains.
- LangChain
- Model Context Protocol (MCP)
- OpenAI API
- Prompt Engineering
Lahyse Belchior · London, UK · Available for Hybrid & Remote
Building Production-Grade AI Agents & Context-Aware RAG Systems
Specializing in LangChain, Pinecone, MCP, and Clinical Guardrails for Digital Health & Enterprise Operations. Based in London — Available for Hybrid & Remote.
- <900ms
- P50 Latency
- 100%
- Deterministic Guardrails
- End-to-End
- Delivery
Core competencies
Tech stack built for production, not demos
Four disciplines that turn a prototype agent into something a clinical or enterprise team can rely on.
Knowledge Retrieval / RAG
Verified retrieval pipelines with citation-aware answers and evaluation-driven chunking.
- Pinecone Vector DB
- OpenAI Embeddings
- Chunking Strategies
Backend & Protocols
Typed, well-documented APIs that integrate cleanly with product frontends and automations.
- FastAPI
- Python
- Webhooks
- REST
- CORS
Observability & Performance
Every run traced and costed, so latency and quality regressions are caught before users notice.
- LangSmith Tracing
- Helicone Latency / Cost Tracking
Featured projects
Case studies with live, testable behaviour
Not slide decks. Try the guardrails yourself in the interactive demo.
Flagship
MindCare AI — Mental Health Triage & Support Agent
Clinical triage agent with deterministic crisis protocols (NHS 111 routing), verified knowledge retrieval, and autonomous tool calling for session scheduling.
- Deterministic crisis protocols with NHS 111 routing — no LLM in the loop for high-risk turns
- Verified knowledge retrieval from NHS and NICE sources with inline citations
- Autonomous MCP tool calling for session scheduling and follow-up reminders
- LangChain
- Pinecone RAG
- MCP
- FastAPI
- Lovable
- LangSmith
Interactive demo preview
MindCare AI · triage agent · demo
Choose a scenario below to see the agent respond.
FintechRequest a walkthrough
Fintech AI Banking & Onboarding Agent
KYC triage, budget analysis, and automated lead routing with human-in-the-loop fallback.
- Lovable
- OpenAI API
- Make / Webhooks
- Airtable
KYC triage that collects, validates and scores onboarding documents
Budget analysis over transaction history with plain-English summaries
Automated lead routing with human-in-the-loop fallback via Make webhooks
System architecture
How a request flows through the stack
Tap any stage to see what happens there and why it exists.
Step 1 User / Lovable UI
The product frontend captures the user turn, attaches session context and streams the response back token by token. Built in Lovable for fast iteration with clinical and product teams.
- Streaming responses
- Session context
- Consent capture
Curriculum vitae
Lahyse Belchior
AI Engineer & Agent Builder | Digital Health & Applied AI. Operating at the intersection of applied LLM systems, cognitive behavioural principles and workflow automation.
Analytical and product-focused engineer experienced in architecting context-aware AI agents, Retrieval-Augmented Generation pipelines and deterministic safety guardrails for digital mental health, healthtech and enterprise operations. Skilled in translating domain-specific protocols into robust architectures using LangChain, MCP, Pinecone and FastAPI, connected to intuitive frontends built with Lovable.
- Digital mental health triage
- CBT logic
- NHS 111 / Samaritans safety overrides
- Workplace wellbeing systems
- EHR data pipelines
- HubSpot CRM workflows
Professional experience
AI Engineer & Product Specialist
2024 – PresentLahybloom Tools · London, UK
- Architected and deployed production-grade conversational AI agents and digital products with LangChain, OpenAI API and Lovable, focused on digital mental health triage and corporate wellbeing.
- Engineered deterministic safety guardrails and crisis detection inside LLM pipelines to flag high-risk clinical keywords and trigger emergency routing (NHS 111 / Samaritans).
- Implemented RAG pipelines backed by Pinecone vector search, grounding responses in validated psychoeducation frameworks and removing hallucination risk in clinical contexts.
- Configured Model Context Protocol (MCP) servers for autonomous tool calling: session scheduling via the Calendly API and automated lead ingestion into CRM databases.
- Integrated LangSmith observability into backend services to monitor P50/P99 latency, trace execution steps and token cost.
Marketing Operations & Automation Consultant
2022 – 2024Lahybloom Tools · London, UK
- Designed data-driven marketing dashboards, automated CRM workflows and lead-generation architectures for UK-based clients.
- Managed multi-platform integrations across HubSpot, Google Ads and custom webhooks to optimise lead flow and conversion tracking.
- Transitioned marketing operations into AI-assisted business infrastructure using Make.com and OpenAI APIs.
More projects
Corporate Mental Health & Burnout Screening Web App
Empathetic assistant for employee burnout screening with a secure analytics dashboard and automated human-in-the-loop escalation.
- Lovable
- OpenAI API
- Cognitive Architecture
- Make.com
Clinical Note & EHR Automation Pipeline
Automated ingestion pipeline turning unstructured clinician session notes into standardised clinical documentation, cutting admin overhead by 30%.
- Make.com
- OpenAI API
- Structured Prompts
- HubSpot / Database
Education & certifications
Associate AI Engineer for Developers
DataCampIn progress
LangChain & Vector Database (Pinecone) RAG Implementation
Hands-on Engineering Track
ChatGPT Prompt Engineering for Developers
DeepLearning.AI
Foundation & Advanced Automation Certification
Make Academy
Inbound Marketing & CRM Workflow Certification
HubSpot Academy
Specialised Studies in Behavioural Analysis & Cognitive Principles
Independent / Domain Expertise
Contact
Ready to deploy intelligent agents for your product?
Tell me about the workflow you want to automate or the knowledge you want to make searchable. I reply to every message personally.