Paravision Lab logoParavision Lab

MVP Portfolio

A collection of GenAI systems built to explore reliable architectures, async workflows, and AI infrastructure patterns.

Discuss the work
Next.jsFastAPIInngestHeyGenSupabaseGemini

Ship ad creatives faster—without trading off brand consistency

An AI pipeline that turns campaign inputs into ready-to-run video ad variants: scripts, voice, avatars, and reliable async rendering.

Designed for retries, idempotency, and clear state tracking across long-running jobs.
Architecture snapshot
Phase A diagram for the AI Video Ads Generator
Read the case study

Queue design, orchestration, and reliability decisions—end to end.

Connect

Production stack (grouped by architecture)

A stable default stack for shipping GenAI products: clear boundaries, predictable orchestration, and operational maturity.

Frontend

Product UI + rendering pipelines.

Product UI + rendering pipelines.
Next.jsRemotionTailwind CSSTypeScript

Backend

APIs, orchestration, background work.

APIs, orchestration, background work.
FastAPIPythonInngest

AI / Media

LLMs, agents, voice, avatars, video.

LLMs, agents, voice, avatars, video.
View stack (9)
LangChainLangGraphGeminiMCPVapiElevenLabsHeyGenD-IDAssembly AI

Infrastructure

Auth + data + deployment primitives.

Auth + data + deployment primitives.
SupabaseClerkArcjetRedisDockerDocker Compose

Documentation in progress

These systems are built and tested; documentation and case studies are being prepared.

Perplexity-style AI Research Tool

Full-stack AI research workspace for iterative investigations.

Agentic routing with web search, SSE streaming, and session persistence.

Web search with source citationsMulti-step reasoning workflowsPersistent research sessionsStructured history and retrieval

AI Short Video Generator

Short-form video pipeline from prompt to rendered MP4.

FastAPI orchestrates generation pipelines with Remotion workers for async rendering.

Deterministic MP4 renderingScript generation workflowsTTS integration for voiceAsync orchestration with state

YouTube Analytics & Growth Tool

Creator analytics and workflow tool for content optimization.

Multi-step analysis workflows using Gemini and YouTube Data API.

Competitor research and analysisSEO insight workflowsCreative planning assistanceAsset workflow support

AI Logo Maker

Branding tool for generating and iterating logo variations.

LangGraph workflows for generation loops with persistent user history.

Prompt guardrails for styleStructured variant scoringPersistent logo historyControlled variation exploration

AI Interview & Scoring Platform

Voice interview platform for structured candidate evaluation.

Real-time voice sessions with stateful orchestration and scoring dashboards.

Low-latency voice interviewsDynamic question generationMulti-dimensional scoringSession history and transcripts

AI Medical Voice Agent (Prototype)

Voice-first medical intake assistant for triage and scheduling.

Real-time clinical conversations with structured data extraction.

Voice intake with STT/TTSMedical data extractionAuthenticated role-based accessAudit-friendly state tracking

Engineering approach

Building AI systems that behave predictably under real workloads—especially async jobs, streaming, and orchestration.

Reliability

Async workflows with retries, idempotency, and explicit failure handling.

Observability

Structured logs, traceable job state, and monitoring for long-running tasks.

Production constraints

Authentication boundaries, data handling, and predictable orchestration.

How I approach AI system development

A structured engineering process for building AI systems involving orchestration, async workflows, and production constraints.

  1. 1

    Problem framing & system design

    Define user workflows, system boundaries, and architecture before implementation. Establish data models and identify risks for complex pipelines.

    Focus areas

    Architecture designData model and API structureWorkflow and orchestration planningRisk identification for complex pipelines
  2. 2

    Implementation & system integration

    Build the core system loop integrating orchestration, persistence, and user-facing workflows. Connect backend services with the UI layer.

    Focus areas

    Agent or workflow orchestrationJob state and persistenceBackend services and APIsUser interface and system interactions
  3. 3

    Evaluation & iteration

    Improve reliability and usability through testing, monitoring, and iterative refinement based on system behavior and output quality.

    Focus areas

    System monitoring and loggingPerformance improvementsEvaluation of model outputsIterative product improvements