I build LLM systems, RAG pipelines, and agentic workflows — from local-first document ingestion engines to multi-agent pipelines coordinated across a single GPU. Background in cybersecurity (CEH, eJPT), which shapes a security-first approach to AI tooling and backend systems.

🌐 Portfolio & case studies: ishaanreddy.in

Tracium — Cross-platform desktop network health monitor (Rust/Tauri). Continuous latency/jitter/loss tracking, speed tests, Wi-Fi diagnostics, bufferbloat grading. ~158MB idle RAM, single-digit-MB headless daemon.

TextStruct — Local-first engine that turns scanned textbooks into structured, RAG-ready data: layout-aware OCR, deterministic math cleanup, pedagogical role classification, hybrid text+vision retrieval over LanceDB. 85–95% role classification accuracy with LLM assistance.

FrameWeaver — Automated multi-agent pipeline generating explanatory videos from a topic: research → planning → image generation (FLUX) + voice synthesis (IndexTTS-2), coordinated to share one GPU's VRAM budget.

LLM Cookbook — Modular, tool-agnostic reference for training language models from scratch: transformers, tokenization, fine-tuning, distillation, scaling laws, and evaluation. 26 topic modules, ~16,900 words.

SLM From Scratch — Training small language models (50M–500M params) from scratch on a single consumer GPU: tokenizers, data pipelines, scaling behavior, and what's actually achievable without a cluster.

Yahoo Finance Dataset — ~81,000 stock tickers scraped across 93 countries and all 11 GICS sectors, with scraper, raw + cleaned data, and company fundamentals.

More project write-ups (architecture, motivation, tradeoffs) at ishaanreddy.in/projects.

AILACDS — Junior AI Developer (April 2026 – Present) Building AI-driven products: AIPathshala, an educational SaaS platform combining RAG-based lesson generation with a document-editor-grade lesson editor, and SalesTaxGST, a custom AI platform for a real Sales Tax/GST consultant with hierarchical RAG over a large document corpus.

MarketsandMarkets — Technology Intern (January 2025 – July 2025) Built an end-to-end Lead Classification System (~89% accuracy). Also shipped an internal FAQ chatbot, a Google-Maps-based vendor discovery tool (GeoSearcher), and a Yahoo Finance data ingestion pipeline.

Grant Thornton Bharat — Cyber Intern (October 2023 – December 2023) Supported enterprise audit, vulnerability assessment, and security analysis workflows.

Telecommunications Consultants India Ltd (TCIL) — Summer Intern (August 2023 – September 2023) Built a Visitor Management System with automation and data workflows.

• LLM integration, RAG systems (hierarchical, hybrid text+vision), agentic workflows
• Prompt orchestration, tool-calling, and backend automation
• Python, PyTorch, Rust, TypeScript, Docker, Linux
• Secure system design and threat-aware automation (CEH, eJPT)

📫 Contact — ishaanreddy.in · LinkedIn