Harinee Mohanakrishnan

About

AI engineer (2+ years; pursuing an online MS at UIUC) who has shipped an AI-powered voice screening platform end-to-end on GCP with strong production monitoring and measurable hiring-process impact (80% reduction in unqualified pass-through; ~50+ hours saved per role). Also built and deployed an AWS-based context-aware hybrid search system using OpenSearch as a vector store, and has hands-on experience with multi-agent LLM orchestration (ReAct) and structured-output guardrails.

Experience

Education

Key Strengths

• Owned and shipped AI voice screening platform end-to-end (Next.js + Django + Celery/Redis) on GCP
• Production observability: Cloud Logging with session-level traceability + Sentry + completion-rate tracking
• Measured business impact: 80% reduction in unqualified interview pass-through; ~50+ recruiter hours saved per role
• Built and deployed hybrid vector search using OpenSearch embeddings for an iOS app backend on AWS (Lambda + S3)
• Designed multi-agent LLM workflow with scoped roles, debate/consensus mechanism, and tool-call orchestration (ReAct)
• Implemented guardrails via system prompts and structured output schema validation to prevent malformed writes
• Led client-driven refactor including OpenAI→Gemini migration; coordinated timeline and engineers; balanced UI vs model/integration tradeoffs
• Uses AI coding tools (Claude Code) to debug complex agent pipeline issues and accelerate re-architecture
• Shipped end-to-end patient-facing voice consultation flow in Next.js App Router + TypeScript
• Strong separation of server vs client concerns (Server Components for sensitive config; client for browser APIs/realtime)
• Performance/UX optimization using Suspense streaming and selective memoization to reduce re-renders
• Production observability ownership (Sentry + Google Analytics) and rapid post-launch bugfixing (mobile audio capture issue)
• Validated data-fetching and security via bundle analysis and Vercel function logs
• Designed Postgres/Django schemas with constraints and indexing; identified and fixed N+1 query issues
• Background job reliability experience with Celery/Redis, including retries and duplicate-prevention for audio saves
• Mentored 2 interns and documented architecture decisions for smoother peer reviews
• Pragmatic vendor trade-off decision-making (Gemini vs OpenAI balancing cost, quality, and GCP integration)