Pratyaksh Krishnna

Pratyaksh Krishnna

AI-Engineer

Hire me

Building AI-native products, from zero to one.

Resume

AI-native software engineer who builds full-stack applications and agentic systems, including multi-agent workflows and RAG pipelines.

I take products from zero to one: I turn ambiguous problems into clear requirements, sound system designs, and production-ready software.

AI speeds up how I build. Product judgment, architecture, code quality, and maintainability stay with me.

Experience

  1. S

    AI Engineer

    Current

    Stealth Mode · Full-time

    Aug 2026 – Present · 2 mos · Remote

  2. H

    Software Engineer Intern

    Hoonartek · Internship

    Jun 2026 – Jul 2026 · 2 mos · Remote

    • Shipped a natural-language analytics agent on BigQuery Conversational Analytics for logistics client XpressBees, so business users could query profitability and lane performance without writing SQL.
    • Built an event-driven ingestion pipeline (GCS → Eventarc → Cloud Run → BigQuery) that auto-loads 6 monthly parquet feeds, with per-file logging and error-folder routing.
    • Scoped the agent to 4 purpose-built semantic views and wrote guardrails that block PII exposure, hallucination and out-of-scope answers.
    • Enforced IAM role-based access and Cloud Logging across the pipeline, and authored the end-to-end SOP now used by the client’s teams.
    GCPBigQueryCloud RunEventarcSQLLLM guardrails

Skills

  • LangChain
  • LangGraph
  • OpenAI
  • Claude
  • Ollama
  • pgvector
  • Supabase
  • FastAPI
  • Celery
  • Zod
  • Python
  • Postman
  • Node
  • Express
  • Bun
  • PostgreSQL
  • MongoDB
  • Redis
  • BullMQ
  • Kafka
  • Socket.IO
  • Prisma
  • Drizzle
  • Docker
  • Nginx
  • AWS
  • GCP
  • BigQuery
  • TypeScript
  • JavaScript
  • React
  • Next
  • Vite
  • TanStack
  • Tailwind
  • shadcn
  • SQL
  • C/C++
  • Rust
  • Linux
  • Git
  • GitHub
  • GitHub Actions

Projects

multi-modal ragLive

multi-modal rag

An agentic RAG system that answers questions grounded in both text and images. Multi-query retrieval over vector search, async document ingestion with background workers, web search fallback, and RAGAS-based evaluation to keep hallucinations in check.

Next.jsPythonFastAPILangChainSupabaseRedisCeleryAWS S3
reclaimOpen Source

reclaim

A bounded multi-agent system for revenue recovery. It detects failed subscription payments and overdue invoices, diagnoses the cause, picks an intervention, executes it within strict policy limits, and proves real recovery against a randomized 10% holdout. Agents reason; deterministic code controls money.

ReactTypeScriptNode.jsExpressPostgreSQLRedisBullMQClaude
View All

Stats

Vagabond - Mountains

“inside, i'm infinite”