Josher Level 1

⦁ Built machine learning and full-stack applications that helped internal teams analyze large datasets, identify operational patterns, and automate decisions previously handled through manual review.
⦁ Developed production data pipelines using Python, SQL, Apache Beam, Spark, and Google Cloud Platform to process large volumes of structured and semi-structured records.
⦁ Trained and deployed classification, ranking, anomaly-detection, and recommendation models using TensorFlow, scikit-learn, and internal ML tools.
⦁ Created feature-generation pipelines that transformed raw event data into reusable training and inference features with validation and lineage tracking.
⦁ Built Python and Java backend services with REST APIs and gRPC endpoints that exposed model predictions to internal applications.
⦁ Developed web dashboards with React and TypeScript that allowed users to review predictions, investigate anomalies, compare trends, and provide feedback.

Led the technical direction of an AI software platform from early product planning through architecture, development, customer onboarding, deployment, and production support.
Built a multi-tenant SaaS platform that allowed business users to connect company data, search internal knowledge, configure AI workflows, and automate repeated operational tasks.
Designed the main platform architecture using Python, FastAPI, Node.js, React, TypeScript, PostgreSQL, Redis, Docker, Kubernetes, and AWS.
Developed AI agents that broke user requests into steps, selected tools, retrieved context, called business APIs, validated outputs, and returned clear responses.
Added human approval steps for sensitive actions so users could review generated content, database updates, and external system changes before execution.
Built a configurable RAG platform that supported document upload, automated chunking, embeddings, metadata filters, hybrid retrieval, reranking, and source attribution.
Designed connectors for CRM systems, cloud storage platforms, databases, email services, and internal APIs, reducing manual data movement between applications.
Created a background job system for long-running AI tasks using message queues, asynchronous workers, retries, idempotency controls, and failure recovery.
Improved response times by adding Redis caching, optimizing PostgreSQL queries, streaming partial results, and moving expensive processing outside the main request path.

⦁ Developed generative AI services and internal tools that helped engineering teams test language-model use cases and move successful prototypes into managed production environments.
⦁ Built reusable APIs for summarization, text generation, information extraction, document classification, semantic search, and question answering.
⦁ Designed RAG pipelines using Amazon S3, OpenSearch, SageMaker, Lambda, ECS, and vector embeddings to answer questions from large internal document collections.
⦁ Created model-serving workflows with versioning, staged deployment, rollback support, autoscaling, request logging, and performance monitoring.
⦁ Built asynchronous processing pipelines for large documents and bulk datasets using queues, serverless functions, object storage, and distributed workers.
⦁ Reduced inference latency by adding request batching, response caching, optimized token limits, parallel retrieval, and task-based model selection.
⦁ Developed evaluation pipelines that measured answer relevance, retrieval accuracy, hallucination rates, latency, cost per request, and formatting compliance.
⦁ Added structured-output validation for workflows requiring consistent JSON responses, extracted fields, classifications, and automated decisions.
⦁ Built internal interfaces with React and TypeScript that allowed users to test prompts, compare model outputs, inspect sources, and review evaluation results.
⦁ Worked with security teams to implement encryption, private networking, access controls, audit logging, and filtering for sensitive information.
⦁ Created cloud environments with Terraform and AWS CloudFormation, allowing development, testing, and production infrastructure to be deployed consistently.
⦁ Added CloudWatch dashboards and alerts for model latency, error rates, queue depth, container capacity, token volume, and service costs.
⦁ Improved reliability in a distributed model-processing environment by strengthening retry logic, checkpointing, job

⦁ Led the technical direction of an AI software platform from early product planning through architecture, development, customer onboarding, deployment, and production support.
⦁ Built a multi-tenant SaaS platform that allowed business users to connect company data, search internal knowledge, configure AI workflows, and automate repeated operational tasks.
⦁ Designed the main platform architecture using Python, FastAPI, Node.js, React, TypeScript, PostgreSQL, Redis, Docker, Kubernetes, and AWS.
⦁ Developed AI agents that broke user requests into steps, selected tools, retrieved context, called business APIs, validated outputs, and returned clear responses.
⦁ Added human approval steps for sensitive actions so users could review generated content, database updates, and external system changes before execution.
⦁ Built a configurable RAG platform that supported document upload, automated chunking, embeddings, metadata filters, hybrid retrieval, reranking, and source attribution.
⦁ Designed connectors for CRM systems, cloud storage platforms, databases, email services, and internal APIs, reducing manual data movement between applications.
⦁ Created a background job system for long-running AI tasks using message queues, asynchronous workers, retries, idempotency controls, and failure recovery.
⦁ Improved response times by adding Redis caching, optimizing PostgreSQL queries, streaming partial results, and moving expensive processing outside the main request path.

⦁ Improved model-serving latency by optimizing preprocessing, reducing network calls, caching repeated features, and moving expensive work out of the request path.
⦁ Designed batch and real-time inference workflows that supported scheduled analysis and low-latency product decisions.
⦁ Added monitoring for feature drift, prediction distributions, data quality, model latency, service errors, and changes in business performance.
⦁ Investigated accuracy degradation caused by changing input data and improved model stability through retraining, updated features, and stronger validation rules.
⦁ Built experiment frameworks that compared candidate models against controlled baselines using precision, recall, accuracy, latency, and downstream user impact.