Open to AI/ML and Gen AI roles

Somala Ajay

Portrait of Somala AjayOpen to work

AI/ML Engineer  ·  Generative AI Engineer  ·  Python Developer

I build LLM applications, RAG pipelines and full-stack AI products, and I run PinGuru, a live Instagram DM automation SaaS. Based in Hyderabad.

View projectsDownload resume
0YEARS EXPERIENCE
0ENTERPRISE PROJECTS
0FAISS VECTORS
0LIVE SAAS

About me

AI/ML and Generative AI engineer with 3+ years of experience building production-grade LLM applications, RAG pipelines and full-stack AI products. My background is in data engineering, data validation and backend API integration, and I have delivered ML/DL models across NLP and computer vision. I work mostly in Python, SQL, FastAPI, LangChain, FAISS and Docker, on GCP, AWS and DigitalOcean.

  • LocationHyderabad, Telangana
  • EducationB.Tech ECE, Aditya University (2018–2022)
  • Live productPinGuru (pinguru.me)
  • Looking forAI/ML, Gen AI and Python roles
Ask Ajay's AIONLINE

Professional experience

AUG 2022 – SEP 2026 · HYDERABAD

Associate Software Engineer at TCS

  • Delivered 3 enterprise projects across ML, deep learning and Generative AI over 4 years, owning data preparation, model development, API development, Docker deployment on AWS and monitoring.
  • Built and maintained Python and FastAPI microservices and REST APIs, with unit testing in pytest and CI/CD pipelines in GitHub Actions.
  • Worked in Agile and Scrum teams with code reviews, technical documentation and regular stakeholder demos.
PROJECT 1 · AUG 2022 – SEP 2023

Visual Defect Detection using Computer Vision

Client: Manufacturing and electronics company
  • Developed a CNN-based image classification model in TensorFlow and Keras, using OpenCV preprocessing and data augmentation to detect surface defects on production-line images. It reached 94% accuracy against 81% for the earlier rule-based check.
  • Used transfer learning (ResNet, MobileNet) and hyperparameter tuning to cut training time by about 40% while improving recall on rare defect classes.
PythonTensorFlowKerasOpenCVCNNScikit-learnFlaskDockerMongoDBGit
PROJECT 2 · OCT 2023 – OCT 2024

Demand Forecasting and Predictive Analytics Platform

Client: Retail and consumer goods company
  • Built time-series forecasting models (LSTM, XGBoost, Scikit-learn regressors) for 1,000+ SKUs with 60+ engineered features such as seasonality, promotions and lag variables, lowering forecast error (MAPE) from 21% to 13%.
  • Created SQL and PostgreSQL data pipelines that processed 5M+ monthly records, and tuned slow queries to cut report generation time by about 45%.
  • Tracked experiments and model versions with MLflow, and added hyperparameter tuning (Optuna) for reproducible training runs.
PythonLSTMXGBoostScikit-learnPyTorchFastAPIPostgreSQLSQLMLflowDockerAWS
PROJECT 3 · NOV 2024 – SEP 2026

Enterprise Knowledge Assistant (RAG and Agentic AI)

Client: Banking and financial services company
  • Built a Retrieval-Augmented Generation (RAG) assistant over 20K+ policy, compliance and process documents using LangChain, a FAISS vector database and sentence-transformer embeddings.
  • Improved retrieval with semantic chunking, metadata filtering, hybrid search (BM25 plus vector) and re-ranking, raising answer relevance by about 30% over the baseline retriever.
  • Developed multi-step agentic workflows with LangGraph and Google ADK, using MCP to connect tools and internal data sources, with tool calling, memory and retry handling.
  • Created an automated LLM evaluation pipeline (retrieval hit rate, faithfulness, answer relevance) to guide prompt and chunking changes.
PythonLangChainLangGraphGoogle ADKMCPRAGFAISSLLMsFastAPIMongoDBDockerAWSGit

Selected work

↗
01 · LIVE PRODUCT

PinGuru

Instagram DM automation SaaS at pinguru.me. Real-time DM triggers, comment-to-DM workflows, multi-tier Stripe billing, JWT and OTP security.

FastAPIMongoDBStripeDigitalOcean
 ↗
02 · RAG

Technical Manual QA System

Production-grade RAG that answers questions from the 270-page NASA Systems Engineering Handbook, with section-level citations.

FastAPIFAISSLlama 3.3Groq
03 · AGENTS

Autonomous Multi-Agent Operations System

A reflex layer handles known triggers. A single agent plans from plain language. For complex goals, a Planner, Executor and Reviewer split the work and self-correct. Containerized FastAPI service with task memory.

GroqDockerSQLitePydantic
04 · DEEP LEARNING

Image Caption Generator

CNN encoder for visual features and an LSTM decoder for caption generation, trained on Flickr8k.

CNNLSTMTensorFlow

Explored events

2026 · Bangalore

AI Quest Hire-athon 2026, i2e Consulting

Took part in the National AI Hackathon and built the Complex Technical Manual QA System: a citation-backed RAG pipeline over the 270-page NASA Systems Engineering Handbook. Indexed 885 semantic vectors in FAISS (all-MiniLM-L6-v2), served Llama-3.3-70B through Groq, and added 0–100% confidence scoring, multi-turn memory, section-level citations and expansion of 20+ NASA acronyms.

FastAPIFAISSGroqsentence-transformersPdfPlumber
At the AI Quest Hire-athon 2026 banner
AI Quest Hire-athon 2026
Participants working on the hackathon floor
The hackathon floor
Working on the project during the hackathon
Building the RAG system

Skills & certifications

  • Generative AI IntroductionMicrosoft Learning
  • Career Essentials in Generative AIMicrosoft and LinkedIn
  • Develop an AI app with the Azure AI Foundry SDKMicrosoft
  • Learning Microsoft 365 Copilot for WorkLinkedIn

Get in touch

I'm open to AI/ML, Gen AI and Python roles in Hyderabad or remote. Email is the fastest way to reach me.

jaydeveloper010@gmail.com+91 76750 10831LinkedInGitHub