Anantha Narayanan
Product Engineer, AI Agents & LLM Systems
I build AI agents that ship to production and keep working when real users show up.
Product Engineer at Oraczen with 2+ years in AI. I design multi-agent systems, RAG pipelines, LLM-as-a-Judge evaluations, and real-time voice agents over WebRTC. I also write about what I build, so other engineers can ship production AI instead of just following tutorials.
Activity
GitHub contributions over the last year.
Projects
Some of the things I've been building.
- 012026
Promptly
AI Prompt Optimization Platform
An AI prompt engineering platform that turns rough prompts into tested, optimized, versioned and reusable production prompts through multi-model collaboration.
LangGraph · FastAPI · Celery · +7
- 02
DebateMind
Memory-Driven AI Debate Trainer
An adaptive AI debate coach that learns how you argue, building a persistent per-user cognitive fingerprint with a knowledge graph and vector memory, and uses it to challenge your real weaknesses over text and voice.
LangGraph · Cognee · Neo4j · +10
Experience
Where I've worked.
Product Engineer · Oraczen
June 2025 – Present
- Engineered an advanced Text-to-SQL chatbot by refactoring a LangGraph application into a ReAct Agent architecture, cutting latency by 40% and raising decision-making accuracy from 40% to 90%.
- Designed a context off-loading mechanism to prevent context degradation in long-running agent sessions, achieving 95% accuracy and a 99% reliability score across extended multi-turn interactions.
- Built an LLM-as-a-Judge evaluation framework for continuous quality monitoring across production AI systems.
- Developed a Deep-Researcher Multi-Agent System where planner and researcher agents generate reports via web search, Text-to-SQL analysis, and image generation (Nano-Banana), with a LongWriter-style plan-then-write pipeline, lifting accuracy from 35% to 90% and automating 80% of manual research.
- Delivered real-time voice agents over WebRTC with tool-calling support, extracting knowledge graphs, actionable items, and summaries from calls, with chat over voice logs via Qdrant-backed long-term memory.
- Established a prompt orchestration layer using XML tag-based structured prompting across the Deep-Researcher and voice agents, isolating untrusted input to harden against prompt injection, with prompt-level guardrails to keep agent behaviour on-policy.
- Built core Zen modules for the company’s Zen Platform, including ZenTools and ContextZen.
Software Engineer · Launch Ventures
July 2024 – June 2025
- Built Wisdom Chat, a RAG application delivering personalized guidance grounded in Bhagavad Gita texts, using a HyDE-based RAG pipeline with LlamaIndex, NeMo Guardrails, and RAGAS-driven quality evaluation.
- Shipped a Flux-powered on-demand image generation feature that visualises abstract spiritual concepts for users.
- Engineered the backend for a Market Research Application in LangGraph with multi-level node classification and validation steps, running web research across Exa, Tavily, and News API and exporting findings as structured CSV reports.
- Implemented a plan-then-write pipeline for report generation, drafting sections independently for coherent long-form output.
- Developed a production-grade async pipeline with FastAPI, Celery, Redis, and Postgres-backed checkpointing, scheduling cron-driven daily runs that email each morning’s research report as a CSV attachment.
Skills
Tools and technologies I use.
- Languages
- PythonC/C++SQLHTML/CSSJavaScriptTypeScript
- AI/ML Frameworks
- LangChainLangGraphLlamaIndexCrewAIGoogle ADKTransformersOpenAI APIGroq
- Web & Protocols
- Vue.jsFastAPITailwind CSSPiniaMCPA2AAG-UIWebRTC
- Databases & Vector Stores
- PostgreSQLClickHouseMySQLPineconeChromaDBQdrantMongoDBNeo4j
- Developer Tools
- Git/GitHubDockerDev ContainersCursorAntigravitypytest
- Libraries
- pandasNumPyMatplotlibRAGASNeMo GuardrailsCustom RAG Pipelines
Contact
Building something with AI agents? I'm open to roles, collaborations, and consulting.