Ingestion & Schema Hygiene
Validate noisy PDFs, DOCX files, and CSV imports upfront using PyMuPDF, python-docx, duplicate checks, and structured parsing before invoking models.
PyMuPDF · Pandas · CSV ChecksGenerative AI Engineer and Python/Django Developer building grounded LLM systems, multi-step agentic workflows, and measurable backend software.
I am a Generative AI and software engineer based in Kollam, Kerala, India, holding a Master of Computer Applications (MCA) from Kerala University (2022–2024, 7.68 CGPA). My engineering career focuses on turning modern language models and automation frameworks into reliable, testable production systems.
Across roles in London (First Quadrant Labs), Trivandrum (Ladder7 Nextstep Solutions LLP), and Dubai (Wadi Al Ain Documents Clearing Services), I have worked across the full AI software lifecycle—from analyzing 57,391 cleaned software reviews with Scikit-Learn, to evaluating multi-document RAG assistants against hallucination and faithfulness benchmarks, to automating multi-source grant monitoring and invoice reminder workflows in Python and Django.
My objective is to build production-oriented Generative AI solutions where prompt design, retrieved context, structured Pydantic/JSON schemas, SQL checks, and automated validation work together seamlessly.
01. Grounded Generation Over Guesswork: Rather than relying on unverified prompts, I anchor LLM responses in retrieved document chunks, embeddings, and strict schema enforcement (Instructor + Pydantic) so downstream systems receive predictable outputs.
02. Evaluation Built Into Delivery: At Ladder7 Nextstep Solutions, I designed evaluation scenarios combining LLM-as-a-Judge checks with embedding similarity, BLEU, and ROUGE to catch hallucinations before deployment.
03. Reliability Across Backend & SQL: From Django REST APIs and SQL consistency tests in Dubai to PostgreSQL Row Level Security in Documint ERP, I verify data integrity, failure handling, and input validation end to end.
Validate noisy PDFs, DOCX files, and CSV imports upfront using PyMuPDF, python-docx, duplicate checks, and structured parsing before invoking models.
PyMuPDF · Pandas · CSV ChecksAnchor generation in vector embeddings and retrieved document chunks, or bind Groq API outputs to strict Pydantic schemas via Instructor.
RAG · Embeddings · PydanticTest faithfulness, relevance, consistency, and hallucination resistance using LLM-as-a-Judge concepts alongside BLEU, ROUGE, and similarity scoring.
LLM-as-a-Judge · BLEU · ROUGEWrap AI logic inside tested Python, Django, or FastAPI REST endpoints backed by SQL validation, Docker containers, and PostgreSQL access policies.
Django · FastAPI · Docker · SQLBuilt multi-document RAG research assistants with retrieved-context grounding, system prompt refinement, structured JSON extraction via Groq API + Pydantic, and evaluation test suites.
RAG · Vector DBs · Groq · InstructorEngineered browser-automated grant monitoring across 5+ funding sources, 3-stage automated invoice reminders, and multi-step orchestration using LangChain, LangGraph, MCP, and A2A concepts.
LangGraph · LangChain · MCP · AutomationDelivered validated Django REST APIs and SQL checks in Dubai alongside NLP sentiment analytics across 57,391 reviews (82.94% logistic regression baseline) for First Quadrant Labs in London.
Python · Django · FastAPI · Scikit-Learn