# AI systems that hold up in production

> Independent GenAI consultancy building production Agentic RAG, multi-agent orchestration and full-stack AI systems, taken end to end from scoping through deployment and handover.

## What this practice builds

- **Agentic RAG**: Hybrid keyword and semantic retrieval, cross-encoder reranking and metadata filtering over a private knowledge base.
- **Multi-agent orchestration**: Stateful, tool-calling agents built with LangChain and LangGraph that plan, act and hold context across long-running workflows.
- **LLM application backends**: FastAPI and Next.js services with typed APIs, structured outputs, streaming and schema validation around the model layer.
- **Document intelligence**: OCR and layout-aware parsing that turns unstructured documents and email into clean, queryable data.
- **Production hardening**: Evaluation, tracing, guardrails, human-in-the-loop review, token budgets and cost control.

## Shipped to production

- **Legal Agentic RAG** (Contracts Specialist Pty Ltd, Australia, 2025): Production RAG over 50,000+ contract and legal PDFs. Hybrid BM25 and dense retrieval with cross-encoder reranking, which measurably reduced the hallucination rate. Stack: FastAPI, LangChain, LangGraph, Pinecone.
- **AI hiring automation** (Cartrends Auto Parts, India, 2025): A seven-stage pipeline from job posting through CV scoring, proctored assessment, async video interview, scheduling and offer letter to onboarding. Cut HR workload by roughly 95% per hire, leaving three human decision points. Stack: FastAPI, PostgreSQL, Next.js 14, Google Gemini, MediaPipe, WebRTC.
- **RFQ to quotation automation** (Under NDA, metals trading, 2026): Email in, draft techno-commercial offer out. Gmail ingestion, structured extraction to a validated JSON schema, supplier matching against an embeddings knowledge base, and a review dashboard. Pre-filtering keeps non-RFQ mail away from the model to cut token spend, and nothing reaches a customer without human approval. Stack: Python, structured outputs, embeddings, human review queue.

## How engagements run

Work is taken end to end and delivered solo: scoping, build, deployment, handover. Every system ships with cost controls, schema validation and a human in the loop, because a demo that impresses is not the same as one a team can run on Monday.

## Contact

- **Book a call:** [30 minutes, free](https://cal.com/shaad-ahmad-rykdvd/30min)
- **Email:** [shaadahmad053@gmail.com](mailto:shaadahmad053@gmail.com)
- **GitHub:** [https://github.com/shaadahmade](https://github.com/shaadahmade)
- **LinkedIn:** [https://www.linkedin.com/in/shaad-a-21423a204](https://www.linkedin.com/in/shaad-a-21423a204)
- **Based in:** Lucknow, India

---

Canonical HTML: https://shaadahmad.com/
Agent brief: https://shaadahmad.com/llms.txt