Skip to content

Swajith Soma

Data Scientist / GenAI Engineer • Fremont, California • s***********@gmail.com • +15******336 • drivetube.ai/•••••

Professional Summary

GenAI Engineer with 5+ years of experience architecting, building, and shipping production AI/ML systems and distributed multi-agent platforms. Specializes in agentic AI, LLM fine-tuning (LoRA, QLoRA, PEFT), RAG pipelines, and GPU-optimized inference. Experienced delivering cloud-native LLMOps and MLOps on Azure, AWS, and GCP, deploying Kubernetes-based GPU clusters, and meeting strict latency SLAs while applying responsible AI practices (GDPR, HIPAA, content safety). Strong Python and FastAPI background with experience in observability, CI/CD, and cross-functional leadership to translate product needs into production GenAI solutions.

Technical Skills

Programming Languages: Python
Web Technologies: REST APIs,GraphQL
Frameworks and Libraries: LangChain,PyTorch,TensorFlow,Hugging Face Transformers,spaCy,BERT,RoBERTa,T5,FastAPI,Flask
Databases: Snowflake,PostgreSQL,MongoDB,Cosmos DB,Prometheus,Grafana
Cloud and DevOps: Azure ML,Azure OpenAI,AKS,AWS SageMaker,EKS,GKE,Kubernetes,Docker,Helm,GitHub Actions,Docker Compose
Testing: PyTest
Data and Analytics: Apache Spark,PySpark,Kafka
Tools and Methodologies: Git
Generative AI & LLMs: LLM fine-tuning,LoRA,QLoRA,PEFT,Prompt engineering,Retrieval-Augmented Generation
Agentic AI & Orchestration: LangGraph,AutoGen,Multi-agent orchestration,MCP servers
Embedding & Vector Search: FAISS,Pinecone,Chroma,Azure AI Search,Re-ranking
MLOps & LLMOps: MLflow,Airflow,Databricks,Model versioning,Drift detection
Inference & GPU Optimization: vLLM,TensorRT,NVIDIA Triton,CUDA,cuDNN

Work Experience

Adobe
San Jose, CA
Data Scientist / GenAI Engineer
Oct 2025 – Present
Worked on generative AI and content-intelligence platform capabilities within Adobe's Experience Cloud, Sensei GenAI, Firefly and Acrobat document workflows.
Tech Stack: Azure OpenAI, AKS, Azure AI Search, Cosmos DB, Llama 2, Mistral, LoRA, QLoRA, PEFT, LangChain, FAISS, Neo4j, vLLM, TensorRT, Prometheus, Grafana, Databricks, MLflow, FastAPI, Kubernetes, GitHub Actions, Python, Apache Kafka, Apache Spark
  • Designed and shipped an enterprise MCP server platform exposing AEP APIs, Azure Cosmos DB, Adobe Analytics streams, and Creative Cloud asset backends as standardized tools for agents, enabling autonomous multi-step workflows and eliminating 10+ weekly manual engineering handoffs.
  • Built a production multi-agent orchestration layer (LangChain Agents + Python dataclass state machine) to decompose tasks, manage memory, and self-correct across content compliance and document intelligence flows, reducing manual content review and campaign assembly effort by 35%+.
  • Implemented LLM fine-tuning pipelines (LoRA, QLoRA, PEFT) on Llama 2 and Mistral using synthetic Adobe-domain data to improve brand compliance classification and semantic extraction, delivering 20%+ accuracy gains and cutting inference compute costs by 30% on Azure GPU fleet.
  • Engineered RAG pipelines integrating Azure AI Search, FAISS, and a Neo4j knowledge graph with hybrid dense + BM25 retrieval and cross-encoder re-ranking, lowering hallucination rates by 40%+ and improving citation-grounded PDF Q&A relevancy for Acrobat AI Assistant.
  • Provisioned AKS-based AI platform with Azure OpenAI, AI Foundry, Event Hubs, Prometheus/Grafana observability and GitHub Actions CI/CD, maintaining 99.5%+ uptime and sub-300ms P95 API latency under high-volume Experience Platform workloads.
  • Optimized NVIDIA GPU training and inference (CUDA, TensorRT quantization, vLLM batching) on NDv4 instances, improving GPU utilization by 30% and reducing generative model inference latency by 40% for real-time creative workloads.
Elevance Health
San Francisco, CA
AI / Machine Learning Engineer
Jun 2024 – Sep 2025
Built ML and NLP solutions for healthcare operations, claims processing, and clinical document retrieval supporting operational and clinical teams.
Tech Stack: AWS SageMaker, EKS, FAISS, BERT, RoBERTa, T5, PyTorch, TensorFlow, FastAPI, MLflow, GitHub Actions, CloudWatch, PostgreSQL, MongoDB, Docker, Python
  • Led demand forecasting platform using ARIMA, SARIMA, and LSTM on SageMaker to model patient volumes and claims seasonality, improving capacity planning accuracy by 18% across 12 regional operations centers.
  • Fine-tuned BERT/RoBERTa/T5 models on internal clinical and claims datasets to extract ICD codes and summarize clinical notes, achieving 92% F1 on clinical NER and reducing manual coding reviews by 40%.
  • Built semantic search over 5M+ clinical documents using transformer embeddings and FAISS, enabling clinicians to find relevant precedents 3x faster versus legacy keyword search.
  • Deployed containerized inference services with FastAPI on EKS, enforcing HIPAA-compliant VPC/IAM controls and MLflow model tracking, sustaining 10,000+ daily inference calls with sub-200ms latency.
  • Automated SageMaker training pipelines with GitHub Actions CI/CD and CloudWatch-based drift monitoring, enabling automated retraining triggers and zero manual pipeline interventions for 9 months.
  • Reduced training time and improved throughput using cuDNN, distributed training and INT8 quantization, cutting training time by 38% and increasing inference throughput by 30% while preserving production accuracy.
Oracle
Mumbai, India
Data Engineer
Feb 2022 – Apr 2023
Engineered high-throughput SaaS telemetry and analytics pipelines on Oracle Cloud to support BI and ML consumers across enterprise Fusion applications.
Tech Stack: Oracle Cloud Infrastructure, Oracle ADW, Apache Spark, Kafka, OCI Data Integration, Python, PySpark, SQL, Informatica, Talend, Tableau, Power BI, Git, Docker
  • Designed Spark and Kafka ingestion pipelines on OCI processing 600GB+ daily SaaS telemetry into analytics-ready datasets, reducing downstream data prep time by 55% for BI and data science teams.
  • Optimized analytical SQL in Oracle ADW using partitioning and materialized views, cutting executive dashboard refresh times from 25 minutes to under 4 minutes.
  • Built Python data quality and lineage frameworks with Oracle Data Catalog to automate schema validation and anomaly detection, catching 250+ data integrity issues before downstream consumption.
  • Implemented Spark-based feature pipelines to supply ML-ready datasets for churn models, contributing to a 22% improvement in churn identification accuracy.
  • Operationalized ETL and integration workflows using OCI Data Integration, Informatica and Talend with CI-driven deployments to ensure repeatable, auditable data pipelines.
  • Created monitoring and alerting for data pipelines and engineered rollback and recovery procedures to minimize data availability incidents for analytics consumers.
Intel
Hyderabad, India
Associate Software Engineer
Aug 2021 – Jan 2022
Supported internal hardware validation engineering by consolidating telemetry and building backend services and dashboards for validation teams.
Tech Stack: Python, FastAPI, React, JavaScript, MySQL, SQLite, PyTest, Git, Jenkins, Linux, Bash
  • Consolidated telemetry from 10+ hardware validation systems into a unified Python REST API with LDAP and RBAC, reducing cross-team data access friction by 50% and providing a single integration endpoint for six teams.
  • Developed interactive React dashboards to show test status, failure trends, and coverage gaps, replacing weekly reports with a live interface used by 150+ engineers daily.
  • Authored PyTest and unittest suites covering 88% of critical code paths, establishing the team's first regression testing baseline and reducing bug escape rate by 42%.
  • Implemented REST API best practices and performance improvements to accelerate telemetry queries and streamline integration with validation automation pipelines.
  • Built backend integrations and data storage using MySQL/SQLite and improved data ingestion reliability and schema management for validation datasets.
  • Documented APIs, onboarding guides, and runbooks to accelerate new team adoption and reduce time-to-productivity for validation engineers.
Docker Associate Software

Projects

CodeLens — Open-Source Code Intelligence Agent | July 2023 – May 2024
Tools Used: LangChain, Claude 2, CodeBERT, tree-sitter, FAISS, NetworkX, FastAPI, React, TypeScript, Redis
  • Built a self-hosted agent to ingest public GitHub repos via GitPython, extract AST nodes with tree-sitter, embed nodes with a contrastively fine-tuned CodeBERT, and index them in FAISS for precise citation-grounded code reasoning.
  • Implemented an orchestration pipeline using LangChain Agents and a Python dataclass state machine (intent routing, retrieval, graph traversal, synthesis) achieving 1.8s P95 end-to-end latency on a 50K-file corpus.
  • Designed a call-graph-aware retrieval using NetworkX to enable cross-file dependency tracing and BFS/DFS impact analysis, improving multi-file architectural reasoning and grounding rates.
  • Delivered SSE/WebSocket FastAPI backend with Celery + Redis for async indexing, embedding cache short-circuiting that cut repeated-query latency by 38%, and a React + Monaco frontend for inline code citations.
  • Benchmarked against GPT-4 and Llama-2-70B on a 150-question evaluation set; achieved 11.4% higher citation accuracy for Claude 2 and 91.2% answer grounding rate with zero hallucinated file references.

Education

Golden Gate University
Master of Science — Business Analytics • San Francisco, CA • May 2023 – Aug 2025
Krishna University
Bachelor of Commerce (Computers) • India • May 2019 – June 2022

Certifications

Microsoft Azure AI Engineer Associate (AI-102) — Microsoft
Microsoft Azure Data Scientist Associate (DP-100) — Microsoft
Applied Data Science with Python Specialization — Coursera
Generative AI with Large Language Models — Coursera

Powered by Drivetube · Create your own profile at drivetube.ai