Horacio Parisotto

GitHub: public repositories with tested code samples (opens in a new tab)

Sr Full-Stack & AI Engineer

Senior Full-Stack & AI Engineer · Founder of OurBlock.io · 6+ years shipping production software, now focused on AI products, performance and cost efficiency.

LLMRAGAI AgentsMCPClaudeOpenAIGeminiVector SearchAI VisionTypeScriptReactReact NativeNext.jsNode.jsGoPythonPostgreSQLMongoDBElasticsearchRedisFirebaseDockerAWSGCPFFmpeg
Production SaaSCanadaVisit site

HarrisTime.ca: Arena Scheduling & Digital Signage

After building venues on the client's legacy arena-scheduling system, I rescued it: traced a runaway Firebase bill to queries loading the entire database history and cut it from $2,353 to $6.49 a month. That fix earned a much bigger mandate: rebuilding the platform from scratch as a multi-venue system driving live digital signage on 30+ screens across 17 facilities for 10 client organizations, including municipalities, across Canada, with an LLM scheduling assistant on Gemini that turns plain-language requests into validated bookings, a field-tested ESP32 horn controller, and automated schedule sync from league websites. Owned it end-to-end across 478 commits and 175 tests gating CI, shipped as Docker images on Google Cloud Run. Migrated incrementally with zero downtime: legacy screens stayed live alongside the new system, and the client has since expanded to Whitby's four arenas and more.

−99.7% Firebase bill$2,353 → $6.49/mo 37k+ events scheduled30+ display screens10 organizations17 facilities600+ teams scheduled175 tests in <1s$0 Firestore/screen
ReactTypeScriptPythonFirebaseLLM (Gemini)IoT (ESP32)

Inside: The Rescue · The Rebuild · In Production

27 pages15 min read

Hiring? Shared only with hiring teams, to protect client confidentiality.

HarrisTime multi-venue dashboard and the Scheda scheduling assistant on a phone
Production MarketplaceChileVisit site

Tur.com: Hybrid Search, Payments & Performance

Built core features of a multi-tenant travel marketplace live in 10 countries, on a custom Go backend with PostgreSQL, MongoDB and AWS. I designed its hybrid semantic search, combining vector and keyword search on Elasticsearch with OpenAI embeddings, so natural phrases find the right tours and the most relevant results come first. It's live in production: empty searches went from 31 to 0 in evaluation, and I validated a later typo-tolerance fix by replaying 3,662 real searches. I also built a self-service backoffice that replaced a developer-dependent workflow, led a performance investigation that raised real Lighthouse scores at release (mobile 43 → 61, desktop 74 → 91), and shipped tiered installment payments. All of it went out across 78 pull requests and 4 repositories, backward-compatible on 45 live pages in Chile, Argentina, Peru, Colombia, Brazil, the United States, Costa Rica, Mexico, Uruguay and Bolivia.

10 countries45 landing pages78 PRs merged31 → 0 empty searches0.41 → 0.75 nDCG@517 DB migrations40+ PRs audited43 → 61 Lighthouse~35 ms cached search
Go BackendVector SearchElasticsearchEmbeddingsRedisPayments

Inside: Landing Redesign & Backoffice · Hybrid Search · Payments & Financing · Performance · Root-Cause Fixes

32 pages18 min read

Hiring? Shared only with hiring teams, to protect client confidentiality.

Tur.com on three phones: activity feed, a tour page and payment methods
Startup MarketplaceArgentinaRead on Forbes

Lit: A Curated Influencer Marketplace

Lead engineer behind Lit, a curated influencer marketplace featured by Forbes. Think Airbnb, but brands list events instead of apartments: creators apply, brands approve them by their audience metrics, a chat opens with the campaign brief, the app validates each creator's QR code at the door, and brands approve the content posted afterwards. I took an unfinished draft from an outside team and built the product end to end, through its launch on Google Play and the App Store, with Mercado Pago plans, a Next.js admin panel for curation and the marketing website.

2 app storesForbes feature49 app screens3 user types9 collaboration states36 client QA items24 app services16 data models4 repositories
React NativeiOS & AndroidTypeScriptFirebaseNext.jsMercado Pago

Inside: The Takeover · The Marketplace · Running the Business · Launch

16 pages9 min read

Hiring? Shared only with hiring teams, to protect client confidentiality.

Lit app screens: sign-up, creator profile, offers feed and Gold plan
Production AI AgentConfidential clientUnited States

Autonomous Remediation Agent for SDLC Compliance

Designed with the CTO of a US platform for SDLC compliance in the AI era. When a deployment gate blocks a non-compliant release, my agent turns the incident into a finished, assigned training plan for the people involved, the same day, closing the loop between compliance failures and the skills that caused them. It acts only through narrow MCP tools under the identity of whoever triggered it, never leaves anything half-created, and says so when it can't produce a safe result instead of inventing one.

Same-day remediation0 human-written steps0 duplicate runs0 standing credentials0 direct DB access4 MCP tools2 execution phases3 autonomy modes3 → 1 intake paths
AI AgentsMCPCompliance (SOC 2 · SOX)Multi-TenantIdempotencyLLM

Inside: The Problem · The Agent · Safety Guarantees · The Platform Underneath

21 pages11 min read

Hiring? Shared only with hiring teams, to protect client confidentiality.

Founder · Production AI PlatformArgentinaVisit site

OurBlock: AI-Powered Security Platform

24/7 AI camera-security platform for residential, commercial and rural properties, built end-to-end as solo founder and live in production on a rural property, with every module running. Every motion event lights up the guard cabin within a second; FFmpeg pulls the frame from the camera's RTSP stream, a vision model separates real threats from false positives, describes each event in one sentence spoken through ElevenLabs and escalates people or vehicles to a critical alert, while a RAG search and a tool-calling agent answer natural-language investigations over the event log. It all runs on a Node.js backend on a single mini PC on site, with no database server.

−94% AI cost/frame94 tests, 100% pass4 AI providers39 API endpoints10 integrations185 commits17.6k lines of TS3-level TTS fallback1 s HLS segments
AI AgentsRAGVector SearchComputer VisionClaudeGemini

Inside: The Product · The AI · Video & Infrastructure · From Simulator to the Field · In the Field

31 pages17 min read

Kept private because it's a live security system protecting a real property. Hiring?

OurBlock website on a laptop and a phone
Production AI Agent · Engineering EnablementConfidential clientLATAM

Automated Code Review Agent

For a client that asked to remain confidential, I turned the team's unwritten review standards into an explicit knowledge base by mining 750+ PR comments across 1,300+ pull requests, then encoded it into a self-gating Claude Code skill that reviews real, open PRs from the team and cross-references sibling code for repeated bugs. On live reviews, the skill does the breadth and I trace what it surfaces down to the root cause.

750+ PR comments1,300+ PRs mined4 repos29 conventionsGo · TS codebases0 auto-posts
Code ReviewTechnical LeadershipAI AgentsClaude CodeEngineering StandardsSafety Constraints

Inside: Mining Team Conventions · The Review Agent · Safety Constraints · Real PR Reviews · A Live Review

11 pages8 min read

Hiring? Shared only with hiring teams, to protect client confidentiality.

Available for senior full-stack & AI roles · Remote or Buenos Aires · Open to contract or full-time

Working with clients in the US, Canada & Chile · English & Spanish

Available for senior full-stack & AI rolesRemote or Buenos Aires · Contract or full-timeClients in the US, Canada & Chile · English & Spanish