P.ROHITH KUMAR VASISTA

P.ROHITH KUMAR VASISTA

Polymath at Heart | Engineering & Building High-Scale Distributed Systems & Agentic AI


About

AI Engineer with 2+ years of production experience architecting stateful multi-agent systems, advanced RAG pipelines, and high-throughput distributed backends for large-scale enterprise environments. At Tata Consultancy Services (Client: Ericsson), I engineer autonomous agentic diagnostic and predictive platforms for Ericsson Network Manager (ENM), governing 5G/4G FCAPS microservices for Tier-1 global telecom operators. My work bridges foundational distributed infrastructure with cutting-edge LLM orchestration, transforming gigabytes of raw, unstructured system telemetry into real-time operational root-cause insights. Core Technical Competencies: Agentic AI & LLM Orchestration: LangGraph (Stateful Multi-Agent DAGs, cyclical fallback routing, parallel fan-out/fan-in), LangChain, LangSmith, Model Context Protocol (MCP), Tool Calling, Multi-Model Routing, Guardrails AI, Magistral-Small, Mistral-Small-24B, Claude Opus/Sonnet, Ollama. RAG & Vector Search: Hybrid Retrieval (Sparse BM25 + Dense Vectors), Qdrant, Milvus, pgvector, ChromaDB, Log-aware chunking, stack-trace semantic segmentation, dynamic metadata filtering, Reranking, Evaluation harnesses (Ragas, TruLens). Distributed Systems & Backend: Python 3.x (AsyncIO, FastAPI, Pydantic), Java, Red Hat JBoss EAP, Infinispan (Embedded/Distributed Caching), JGroups, GossipRouter, JMS messaging, REST APIs, Microservices, PostgreSQL, Clean Architecture. MLOps & Infrastructure: MLflow, Docker, Kubernetes, Helm, GitHub Actions CI/CD, Prometheus, Grafana, OpenTelemetry, Drift Detection, A/B Testing. Track Record of Impact: At Ericsson, I architected an autonomous multi-agent log diagnostic engine that reduced Mean Time to Diagnose (MTTD) by >70% and cut L1 triage noise by 85% across 24 GB multi-node daily telemetry bundles. Simultaneously, I designed a multi-agent KPI analytics platform to compress 72-hour continuous load-test KPI evaluation cycles from days to minutes.

Skills
Agentic AI
Multi-Agent Systems
LangGraph
LLMs
Guarded RAG
AI-MLOps
Model Context Protocol (MCP)
MLflow
Vector Databases
Python
Java
FastAPI
System Design
Distributed Systems
Infinispan
JBoss EAP
JGroups
Hibernate
PostgreSQL
DevOPS
Kubernetes
LangChain
CI/CD
Linux
Experience

Software Developer (DEV 2)

TCS (Ericsson)

Jan 2026 to Present
  • Autonomous Incident Analysis & Log Diagnostic Engine (Agentic AI)
  • - Engineered a stateful multi-agent orchestration workflow using LangGraph and LangChain, deploying Magistral-Small for deep root-cause reasoning and Mistral-Small-24B for structured executive dashboard reporting.
  • - Implemented a Hybrid RAG pipeline combining sparse BM25 keyword search with dense vector embeddings backed by Qdrant and Milvus, incorporating log-structure-aware sliding window chunking, stack-trace semantic segmentation, and dynamic 15-day metadata filtering.
  • - Integrated LangSmith for real-time prompt tracing, agent trajectory debugging, context recall tracking, and continuous evaluation across RAG retrieval accuracy and model groundedness.
  • - Reduced Mean Time to Diagnose (MTTD), reduced L1 log analysis noise, and achieved high Top-3 Root Cause Analysis (Top-3 RCA) precision across high-load release testing environments.
  • - Third-Party Product (3PP) Governance & Leadership: Served as the designated 3PP Champion, spearheading technical evaluations, constraint analysis, and establishing architectural best practices to guide engineering teams on optimal third-party tool integration.
  • - Workflow Architecture & Management: Owned and governed complete enterprise workflow systems using Camunda BPM across Ericsson Network Manager (ENM) and EO products, ensuring seamless process automation and high availability.
  • - License Control & System Ownership: Managed the end-to-end License Control Monitoring (LCM) component, handling over 500+ licenses across core ENM enterprise applications and telecommunication nodes.
  • - Performance Tuning & Memory Optimization: Engineered a critical memory-based dimensional architecture change for LCM, boosting system efficiency and response time while resolving Out-of-Memory (OOM) crashes and service restarts.

Software Developer (DEV 1)

TCS (Ericsson)

Aug 2024 to Jan 2026
  • Multi-Agent Telecom KPI Evolution & Predictive Analytics Platform:
  • - Developed a DAG-based multi-agent pipeline in LangGraph (Ingestion Agent, Noise-Filtering Agent, Release Comparator, Predictive Reasoner) running parallel fan-out/fan-in evaluation.
  • - Integrated Model Context Protocol (MCP) tool calls against Jira and Confluence issue registries to automatically cross-reference anomalies against known flakiness and past bug histories.
  • - Utilized Magistral-Small to run non-linear regression on microservice resource utilization trends and Mistral-Small-24B to synthesize macro architectural evolution reports and capacity forecasts.
  • - Accelerated release-readiness evaluation cycles from days to minutes, achieved ultra-low MAPE on microservice degradation predictions, and delivered proactive capacity recommendations for Tier-1 telecom carrier deployments.
  • Core Platform Engineering — ENM SDK Cache & Clustering Modules
  • - Delivered an Infinispan embedded-cache system serving 30+ production services, 200+ named caches, and 80+ clusters, following comparative proof-of-concept (POC) evaluations against Hazelcast and Apache Ignite.
  • - Built high-availability clustering on JBoss EAP using JGroups, GossipRouter, and JMS for cache coordination and asynchronous event-driven messaging.
  • - Integrated Prometheus metrics exporters for distributed cache hit rates, eviction thresholds, and cluster heartbeats, actively leading architectural code reviews.
  • - Achieved a ~50% cache performance improvement across the ENM platform and guaranteed zero split-brain partition failures during failover testing.
Projects

DocuSage AI Platform

Architected and developed a secure local document workspace utilizing Docker Compose and Ollama to parse and summarize PDF, DOCX, TXT, and Markdown files entirely on-premise without requiring paid cloud LLM APIs. Implemented robust account authentication and asynchronous document processing workflows (Uploaded -->> Processing -->> Ready or Failed) featuring automated retry mechanisms, trash management, and a strict 1 GB per-account limit. Configured an offline evaluation harness and integrated automated continuous integration workflows via GitHub Actions.

Ollama (LLM)
GaurdedRAG
Python
FastAPI
React
Docker

MLOps House Price Platform

Engineered a complete training-to-serving machine learning pipeline that automatically compares four model candidates per run on versioned datasets. Integrated MLflow for experiment tracking, metric logging, model registry management, and the automated promotion of production @champion models. Containerized and served the registered model through a production-ready, Dockerized FastAPI microservice.

Python
MLflow
FastAPI
Docker
Education

M.Tech in Artificial Intelligence and Machine Learning

BITS Pilani, Hyderabad

Apr 2026 to Sept 2028

B.Tech in Computer Science Engineering

VNR Vignana Jyothi Institute of Engineering and Technology

Sept 2020 to May 2024