LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Turn Chaos Into Structure. A Type-Safe AI Agent that extracts valid JSON from unstructured data using PydanticAI, FastHTML, and Gemini 2.5.
Universal page-to-data extractor, turn URLs or HTML into clean Markdown, JSON, or Text for RAG and AI workflows. Rust API + Tauri desktop app.
Universal prompt library for structured outputs & ready-to-use content. Teachers get lesson plans, developers get reliable JSON/CSV. Works across GPT-4, Claude, Gemini.
n8n workflow templates extractor
Fine-tune Qwen3-0.6B for resume parsing using LoRA
A full-stack AI Document Intelligence API that extracts, parses, and converts unstructured PDF content into summaries, key insights, and structured data outputs using FastAPI, PyMuPDF, and Google Gemini API.
LLM extraction of 5 CRM fields from Japanese meeting notes (Qwen 2.5 72B) with exact-match eval and prompt-injection defenses. 60% exact match on 30 synthetic notes; attack success 45%→6.7% (self-authored set). Streamlit + Docker.
Schema-validated LLM extraction eval harness (Anthropic Claude): failure taxonomy (correct/missing/wrong/hallucinated), per-field accuracy, deterministic offline replay, cost/latency accounting, and confidence-gated human review. Structured outputs over grid-interconnection docs as the worked example.
Structured JSON extraction from LLMs with validation, repair, and streaming.
AI-powered structured web scraper that visually builds JSON schemas and uses Gemini 2.5 Flash & Playwright to extract clean, validated JSON with >80% DOM noise reduction.
Turn handwritten forms, notes, and scanned paperwork into automation-ready JSON
Production-style fine-tuning project for schema-constrained JSON extraction using QLoRA + DPO, with reproducible evals, training curves, and vLLM benchmarks.
Zero-shot vs few-shot JSON-extraction agent for finance complaint triage (Qwen2.5-7B on Kaggle GPU, Groq fallback)
A Json Analysis Tool
Professional-grade AI logistics pipeline built with Java 17 and Spring Boot 3. Converts unstructured documents into validated JSON via non-blocking WebClient adapters (Groq/Llama 3.1). Features real-time dashboard, Slack/Email notifications, and secure error handling."
Fine-tuned Qwen2.5-7B on Fireworks AI for structured JSON extraction from job postings. LoRA SFT + DPO | FastAPI | +47% over baseline.
Hybrid deterministic + distilled LLM structured extraction (JSON): schema registry, hard-example generation, teacher-validation gate, knowledge distillation, hybrid routing. V2/V3 research whitepapers, full provenance, evidence graphs.
End-to-end LLM alignment pipeline (SFT + DPO) using QLoRA, Axolotl, and Hugging Face TRL to fine-tune Qwen 2.5 8B for strict JSON extraction and calibrated tool-calling.
Udgam-384M: a 384M-parameter English model for on-device tool calling, trained from scratch. Docs, runnable examples and checksums; weights on Hugging Face.