Kathan Modh — Applied AI & NLP Engineer

Software Development Consultant — Altum Labs

Jun 2026 — Present · Ahmedabad, India

• Architected and shipped a production full-stack system (IT Asset Management, deployed on Azure and used daily) on a 6-layer architecture using .NET, Angular, EF Core and SQL Server, secured with JWT + Azure AD.
• Built and evaluated production LLM and agentic AI features on LangChain + MCP — a data-extraction pipeline reaching 95%+ accuracy across 500+ sources, plus OCR and chatbot microservices — with an evaluation harness for reliability.
• Wrote 125 automated tests (xUnit + pytest) with Playwright and CI/CD, and reduced batch inference latency 12.8% and manual verification effort 60% through optimization and validation.

95%+ extraction accuracy · 125 automated tests · −12.8% inference latency · −60% manual verification

Stack: .NET, Angular, EF Core, SQL Server, LangChain, MCP, Azure, Playwright

Undergraduate Researcher — CHARUSAT · UGSF Fellowship

2025 — Present · Anand, India

• Contributing to PhD-level research on fake news detection under Prof. H. Thakkar and Dr. A. Thakkar — experimental evaluation on 50K+ labeled samples, engineered NLP features (TF-IDF, sentiment, linguistic cues), and ablation studies on classification F1.

50K+ labeled samples

Stack: NLP, TF-IDF, Sentiment, Ablation studies

Software Developer Intern — WORK24 (Raidlayer Webhost)

May — Jun 2025 · Ahmedabad, India

• Built a Selenium web scraper extracting Google Cloud usage data for a data-analytics client, handling proxy and anti-bot blocking through low-level request handling — tested across many websites.
• Delivered a RAG-powered document chatbot for a Dubai law firm, and automated backend workflows with Python/Flask, raising system uptime from 94% to 99%.

99% uptime, up from 94%

Stack: Python, Flask, Selenium, RAG, Automation

LatentGate — Open source

LLM optimization pipeline, published on PyPI with an MCP server. ~80% llm api cost cut.

Stack: Python, FastAPI, MCP

DevMirror — Product

Real-time collaborative code editor with Redis pub/sub for distributed, low-latency sync. 10+ languages supported.

Stack: React, Node.js, Redis, WebSocket

ExamGenie — Research

Multilingual question generation. Extractive and abstractive transformer pipelines, fine-tuning T5 and BERT-based models across three languages. Targeting EMNLP 2027. BLEU 0.42, ROUGE-L 0.61, Languages 3.