Kelvin Biffi

Summary

Software engineer with 15+ years shipping production systems, most of it on software that carries money and identity: an authentication gateway on Keycloak with MFA serving millions of users, subscription billing on Stripe, payment idempotency on WorldPay, and a US healthcare data platform with per-client access control.

For the last few years that engineering has been pointed at LLM systems: agents with tool calling, MCP servers, retrieval pipelines, and the validation layer that keeps them from lying. An agent does not fail loudly. It returns something plausible, the logs look clean, and weeks later someone finds a corrupted record. That reliability layer is the same discipline checkout and authentication already demanded. I also build the interface on top: the approval card, the refusal state, the citation and the cost meter that decide whether a person trusts the system.

Experience

AI Automation Engineer, Hire With Near (US recruiting platform) · May 2026 - Aug 2026

• Built Near Companion end to end, a meeting assistant for recruiting: a Chrome extension on top of an LLM pipeline in Python and TypeScript, integrating Recall.ai, Supabase, HubSpot, Cloudflare R2 and n8n.
• A bot joins the intake call as a silent participant and the live transcript feeds generation while the call is still happening. By the time the call ends, the structured job description is written and the matching candidates are ranked.
• Every model output passes schema validation and field-level assertions before it can touch a CRM record. Invalid output is retried with a corrective prompt, then falls back to a flagged draft for human review instead of writing partial data.
• Candidate discovery, matching and ranking with contextual retrieval, explaining why each candidate fits. Shareable result pages served from Cloudflare R2.

Senior Full Stack Engineer, Together Devs (payments platform) · Apr 2026 - Aug 2026

• Built the payment integration connecting WorldPay to the crypto trade and transfer side of the product.
• Implemented idempotency across the payment endpoints, so a retried request or a duplicated webhook cannot charge twice or trigger a second transfer.
• Worked the failure paths that decide whether money actually moves correctly: retries, out of order callbacks, and consistency between the acquirer and internal records.

Senior AI Engineer, 8 Figure Agency (US) · Feb 2025 - Apr 2026

• Multi-agent systems with tool and API integrations for full-funnel marketing (capture, enrichment, CRM, outreach), using explicit orchestration rather than a framework so every tool call stays inspectable.
• Guardrails on irreversible actions: no agent writes to a CRM, sends outreach or spends budget without passing validation first, and anything below threshold stops for a human.
• n8n pipelines across Gmail, GoHighLevel, Apollo, Datagma, Instantly and PhantomBuster. Meta Ads to BigQuery pipeline end to end (API, backfill, schema alignment with Airtable COGS data).
• Owned the Lead Magnet SaaS: ebook generation from viral content, iterating output quality to remove AI slop, Typeform quiz integration and Beehiiv subscriber automation.

Senior Software Engineer, Firstleaf (US wine subscription e-commerce) · Jan 2021 - Feb 2026

• Subscription Experience team: subscription, billing and account management flows on Stripe with webhook-driven lifecycle events, used by thousands of paying members.
• Landing systems and A/B experimentation at scale with Instapage and Plasmic, GraphQL through Gatsby, Shopify integration, and the launch of COLA Wines, a sister product line.
• LLM-driven SEO content platform, grounded in structured product data and validated before publish. Semantic and sentiment analysis of customer feedback turned into input for product decisions, before the LLM wave.

Tech Lead, Zallpy / Surgo Health (US healthcare data platform) · Aug 2024 - Apr 2025

• Led the build and maintenance of a health data research and analytics platform sold to pharmaceutical companies, in Next.js, Express and PostgreSQL.
• Tiered data access and granularity per client, so each client sees only the level of detail their agreement allows.
• Geographic analysis layer in Leaflet with mobility barrier data, so researchers read health indicators over territory instead of in a table.

Senior Software Engineer, Zartis (US and Dubai enterprise clients) · Dec 2021 - Dec 2024

• DIFC, Dubai International Financial Centre: compliance-grade frontend flows where the correctness of what is displayed is a regulatory concern.
• 605 (US media analytics): analytics frontends for high-volume data visualization in React and Node.js.
• Knowledge Gap: LLM over client documentation that surfaces gaps and contradictions across the corpus. Product engineering for a US real estate SaaS client.

Senior Frontend Engineer, Zallpy / Lojas Renner (national retail chain) · Jul 2022 - May 2023

• Led the enterprise login and registration system on the identity and access management squad. Keycloak with MFA, an authentication gateway used by millions of customers, where a regression means a locked out customer.

Senior Front-End Developer Analyst, Pmweb · Apr 2018 - May 2021

• Led a modular Vue.js landing page builder that let marketers create pages on their own, cutting development effort by over 70%.
• CRO projects through A/B testing and analytics, internal tooling to remove manual work, technical lead on an internal platform redesign, mentoring junior developers.

Front-End Developer, OSF Global Services · Sep 2017 - Dec 2017

• Frontend for e-commerce on Salesforce Commerce Cloud, with international teams.

E-commerce Front-End Developer, Object Edge · Dec 2015 - Jan 2017

• Responsive, mobile-first interfaces for Oracle Commerce Cloud projects, with design-system-aligned components.

Backend Developer, Unilasalle · Feb 2013 - Nov 2015

• PHP for academic and finance systems, SOAP web services, Oracle and SQL Server integrations, and an exam simulation system.

Products i built and run

ViralRecast · 2026 · viralrecast.com

• Finds what is going viral on TikTok, Reels and YouTube, explains why each post worked and cuts the vertical clips with captions. Web dashboard plus an MCP server with OAuth 2.1, PKCE and dynamic client registration for Claude and ChatGPT.
• Postgres 16 with pgvector and HNSW for similarity, an atomic credit wallet that refunds failed jobs, Stripe. React, TypeScript, PHP 8.3, Redis, OpenRouter.

Renderster · 2026 · renderster.com

• Video rendering as a service and the media engine behind ViralRecast. Go 1.23 API and workers on an asynq queue with retry, backoff, visibility timeout and dead-letter, FFmpeg, output on Cloudflare R2, OpenAPI and llms.txt generated from code.

Mentezinha · May 2025 - present · mentezinha.app

• Child-safe conversational AI on Google Vertex AI with retrieval per child on managed vector search, parental governance, real-time supervision and a guardrail with the authority to refuse. React, TypeScript, Next.js, Node.js, Gemini.

Linkezera · Jul 2025 - present · linkezera.com

• Link and bio SaaS with 3,700+ users across LATAM, built end to end: React and PHP/MySQL, subscription billing and gated features, real-time analytics (CTR, geo, device, referrer), custom design system, MCP access to quizzes and leads.

StartAtende · startatende.com

• AI customer support on WhatsApp with intent routing, cost per conversation, and sentiment and product-gap analysis over the agent's own conversations, exposed over MCP.

SunoBoard · sunoboard.com

• AI music studio with an open-source MCP server on npm (`@yellowkode/suno-mcp`), built to score my own games.

MentalKare · Nov 2020 - present · mentalkare.org

• Non-profit mental health platform with verified, multilingual content for families navigating neurodiversity.

YellowKode Academy, Founder and Lead Instructor · Oct 2024 - present · yellowkode.com

• Workshops and courses on AI, prompt engineering and automation with n8n and LLMs. Execution over theory.

K3 Company, software consultancy · May 2021 - present

Open source

• ollos-mcp (npm): local Whisper transcription, speaker segmentation, keyframes and OCR for agents, in Node through ONNX Runtime, with confidence per segment.
• suno-mcp and suno-mcp-core (npm) · yk-agent and yk-chat, local agents with no API key · squadron, a skeleton for one person and a team of agents · Guide-Book (npm `guide-book-generator`), a design system generator.

Stack

Languages Python, TypeScript, JavaScript, Go, SQL, PHP · Frontend React, Next.js, Gatsby, Vite, Tailwind, Chrome Extensions, design systems · Backend Node.js, Express, FastAPI, REST, GraphQL, Apollo, webhooks, Redis, asynq, Stripe · Data PostgreSQL, pgvector, MySQL, MongoDB, Supabase, BigQuery · AI agents, multi-agent, tool calling, MCP servers and clients, RAG, managed vector search, Vertex AI, Gemini, OpenAI, Claude API, guardrails, LLM validation and evaluation, WhisperX, n8n · Auth OAuth 2.1, PKCE, Keycloak, MFA · Ops Docker, CI/CD, GitHub Actions, Cloudflare R2, CloudWatch · Testing Vitest, Cypress, Playwright

Education

MBA, Artificial Intelligence, Faculdade Exame (2025 - 2026, in progress) · MBA, Market Intelligence · Postgraduate, Leadership and Coaching · BSc, Systems Analysis and Development

Books

Almost thirty published books. Four are the exact stack I work in: MCP na Prática com Python, Context Engineering: O Guia Definitivo para LLMs, Agentes de IA com Python and Spec Driven Development na Prática.

Languages

Portuguese (native) · English (C1 Advanced, EF SET 66/100, cert.efset.org/jx9FTN)