Arun Kumar

AI/ML Engineer with around 2.5 years of professional experience building Generative AI and agentic systems on Python backends. I design and ship RAG pipelines, multi-agent workflows, and LLM-integrated applications — deployed with Docker on AWS, not left as demos. I also own Scripten, a multi-agent content platform I built end to end, from system design and implementation through testing and AWS deployment.

My focus is the practical side of LLM systems: retrieval quality, latency, cost per request, and error handling.

AI/ML Engineer — Generative AI & Backend Systems at Aryu Enterprises Pvt. Ltd. (2025-01 – Present)

Design and ship LLM-backed systems on a shared FastAPI/Django backend: a RAG knowledge agent over internal HR documents, multi-agent orchestration pipelines, a real-time voice agent, and a document-parsing service. Work on a 3-member Python team — running sprint planning, code reviews, and coordinating delivery with the digital marketing team. Deploy services with Docker and CI/CD on AWS; apply standard API security (input validation, role-based access, rate limiting) across endpoints.

• Built a token-throttling and model-routing layer that controls prompt context-window usage across requests, lowering per-request inference cost without breaking structured outputs.
• Removed N+1 query patterns across PostgreSQL-backed services using select_related/prefetch_related and composite indexes, with assertNumQueries regression checks keeping API responses under 100 ms.
• Institutional HRMS Knowledge Agent (RAG pipeline, FastAPI, ChromaDB): Built the full RAG pipeline over unstructured HRMS documents — chunking, embedding, semantic retrieval, re-ranking — so admin queries are answered from private internal data instead of manual lookups; query resolution went from hours to seconds. Added an LLM orchestration layer with guardrails for multi-step admin queries, with no external data exposure.
• Resume Parsing Engine (FastAPI, Gemini, Groq, Llama Parser): Built async document-parsing workers that convert unstructured resumes into structured company templates using Llama Parser plus LLM-based field mapping. Integrated a Gemini + Groq fallback so parsing keeps running through API quota limits or latency spikes, and routed/throttled prompts per request to control inference cost while preserving extraction accuracy. Secured endpoints with role-based access and input validation, segmented by organizational hierarchy.
• Autonomous Voice Agent — VoIP Tele-Calling (FastAPI WebSockets, Sarvam AI, Groq, Vobiz): Replaced manual CRM cold-calling with an autonomous tele-calling agent — Sarvam AI STT, Groq LLM reasoning, and Vobiz TTS in a real-time async pipeline over WebSockets. Tuned per-stage latency budgets (STT 120 ms, LLM 150 ms, TTS 130 ms) with first-sentence token chunking and audio pre-buffering, keeping end-to-end audio latency under 500 ms. Used Redis for VoIP session state, sequence-ID audio buffer tracking, and graceful WebSocket reconnects, so calls survive telephony instability without losing mid-call context.
• Enterprise LMS & CRM Platform (Django REST Framework, PostgreSQL, React Native): Built an LMS and lead CRM on Django REST Framework with multi-tier RBAC and rate limiting, keeping the API layer fast through query optimization and composite index tuning.

Integrated

Meta's WhatsApp Cloud API into the CRM — webhook-based inbound handling and delivery/read tracking — so staff send and receive lead messages inside the application, with conversation history linked to lead records.

Added

Kafka and Celery for batch and background processing of messaging and campaign workloads. Currently extending the messaging module with WhatsApp template management and segmented broadcast campaigns (in progress).

Relationship Manager at City Union Bank (2023-06 – 2024-07)

Managed banking relationships for retail customers — account servicing, product-related queries, and day-to-day support at the branch level. Handled financial products and customer portfolio follow-ups in a target-driven environment; left with a working knowledge of banking operations that now informs finance-domain AI work.

MBA in Operations & Finance – Rathinam Institute of Management (2021-01 – 2023-12)

Bachelor of Computer Applications – Dharumapuram Adheenam Arts and Science College (2018-01 – 2021-12)