S
Full-stack developer & AI engineer · 2025 — now
SupplierVerse.ai
AI procurement intelligence for suppliers and public buyers.
A B2B platform that helps suppliers find relevant public-sector opportunities, decide what is worth bidding for, and draft evidence-grounded responses, with a person in control of every answer that leaves the system.
LangGraph · RAG · Azure OpenAI
Multi-step AI bid writingSemantic opportunity matchingHuman-approved output
What I built
- Worked across the whole product: Next.js workspaces and Server Actions on the front, a Python FastAPI service for AI and compute behind it, and PostgreSQL underneath.
- Built an agentic bid-writing workflow on LangGraph and Azure OpenAI: a lead step plans the answer, retrieves the supplier's own evidence, drafts, critiques from an evaluator's point of view and revises, with typed structured output and live progress over SSE.
- Designed the guardrails that make the output trustworthy: evidence allow-lists, claim provenance, explicit ‘needs input’ markers instead of invented facts, bounded loops and human approval before anything is used.
- Worked on retrieval and matching: ingestion of public tender data into staged, recoverable pipelines, embedding-based semantic search combined with structured signals, and LLM-written match explanations.
- Built tool-calling assistants that are always scoped server-side to the signed-in organisation, plus a second graph for bid-readiness assessment.
- Delivered collaboration (comments, mentions, tasks, attachments), Stripe billing and lifecycle email, and shared identity work with role claims and row-level security.
Product surface
- Opportunity discovery and matching for suppliers
- AI-assisted bid and capability-question drafting
- Bid readiness and eligibility assessment
- Team collaboration on opportunities and tasks
- Subscription billing and lifecycle email
Techniques & stack
LangGraphLangChainAzure OpenAIRAGEmbeddingspgvectorFastAPINext.jsPostgreSQLSSE
A note on confidentiality
This is employer-owned work. I describe what I built and the techniques I used, not internal architecture, data sources, volumes or roadmaps. Happy to go deeper in an interview within those limits.