# Afraz Ahmed

> AI Engineer building production LLM and agentic systems for enterprise clients across RAG pipelines, multi-agent architectures, and voice-driven conversational AI on Azure and AWS.

Contact: afraz3301@gmail.com | LinkedIn: https://linkedin.com/in/afraz03 | GitHub: https://github.com/Afraz33

## Summary

AI Engineer building production LLM and agentic systems for enterprise clients across RAG pipelines, multi-agent architectures, and voice-driven conversational AI on Azure and AWS. At Applivity, owns the architecture for Incident AI, Inspection AI, and a video analysis pipeline for training content, plus a dual-agent document intelligence system that cuts hallucination rates and manual review time. Works across LangChain, LangGraph, Azure AI Search, Pinecone, FAISS, BM25, and prompt engineering to ship LLM products end to end, backed by a Python, Node.js, and React full stack. AWS Certified Cloud Practitioner.

## Skills

- Generative AI & LLM Engineering: OpenAI API, Azure OpenAI (GPT-4, GPT-4.1, GPT-5 mini), RAG, LangChain, LangGraph, LlamaIndex, embeddings, prompt engineering (few-shot, chain of thought, structured output constraints), LLM-as-judge evaluation, model evaluation and benchmarking
- Agentic AI & Automation: multi-agent architectures (Extraction/Verification pattern), AI agents, chatbots, copilots, tool calling, workflow orchestration, human-in-the-loop review, MCP integrations, n8n
- Machine Learning & NLP: Python, TensorFlow, PyTorch, Scikit-learn, NLP, deep learning, structured/unstructured data evaluation
- Search & Retrieval: Azure AI Search (vector/hybrid), Pinecone, FAISS, ChromaDB, BM25, fuzzy search, semantic search, chunking/indexing
- Voice & Conversational AI: Azure Speech SDK (STT/TTS), sherpa-onnx (Whisper, Piper), on-device/offline speech pipelines, multilingual (9+ languages)
- Document Intelligence: Azure Document Intelligence (Form Recognizer), unstructured-to-structured extraction, invoice/form/report parsing
- Backend & APIs: Node.js, Python, REST API design, JWT, OAuth2, OpenID Connect, microservices
- Cloud & Data: Azure (App Service, OpenAI, AI Search, Document Intelligence, Speech), AWS (Lambda, ECS, S3, RDS, IAM, EventBridge, SQS), GCP, Neo4j, multi-cloud pipelines

## Featured Project

### Engineering Memory OS: CockroachDB × AWS Hackathon (Aug 2026)
Leading a 2-3 person team building an AI-powered institutional memory platform integrating GitHub, GitLab, Slack, and Jira to capture engineering decisions and answer natural-language questions about why decisions were made, using persistent AI memory. Owns the retrieval/memory layer architecture on CockroachDB, a RAG-based query engine, and MCP-based connectors. Scope spans LLM/agent design, vector retrieval, AWS deployment, and full-stack delivery.

## Experience

### Software Engineer, Applivity (Aug 2024 to Present)

**Inspection AI | RealWear First Agentic Field Inspection Platform**
Owns architecture for an agentic, voice-driven inspection platform for RealWear devices (.NET Core backend, React/TypeScript frontend, Azure App Service, GitHub Actions CI/CD). Indexed inspection manuals in Azure AI Search combining BM25, fuzzy search, and vector similarity for hands-free technician queries. Applied prompt engineering for voice-ready answers. Set JWT/RBAC auth model with OAuth2/OIDC. Led an offline, on-device speech pipeline for RealWear hardware (Kotlin, Jetpack Compose, sherpa-onnx running Whisper tiny.en and tiered Piper TTS voices locally) with tier-aware model selection and graceful degradation.

**Video Analysis AI | Training Content Search Pipeline**
Built a video analysis pipeline chunking training footage into indexed, timestamped segments, searchable via Azure AI Search and Pinecone (keyword + semantic), tuned for training-specific chunking/labeling precision.

**Incident AI | Voice Driven Field Incident Reporting**
Led delivery of a production voice-driven incident reporting system for field engineers in remote oil and gas environments (ADNOC Drilling) using Azure Speech SDK across 9+ languages. Cut incident logging time from hours to under 2 minutes.

**PDF & Document Intelligence Pipeline | Multi-Agent AI Architecture**
Designed a dual-agent system: an Extraction Agent (prompt-engineered LLM calls) and a Verification Agent (cross-validates against business rules, flags anomalies). Built on Azure Document Intelligence with few-shot and chain-of-thought prompting to reduce hallucination rates.

**Rapticore | Cloud Security & Data Governance SaaS Platform**
Designed multi-cloud (AWS/GCP/Azure) data acquisition pipelines feeding an automated data governance catalog for AI-driven risk scoring. Built a Neo4j graph database + FastAPI REST APIs, 95% faster than relational approaches for lineage queries. Built a compliance rule engine (CIS AWS Benchmark v1.4, 50+ checks). Event-driven ingestion (EventBridge/SQS) cutting detection latency from hours to under 10 seconds. Directed containerized microservices (Docker, ECS) with Terraform/CDK, reducing onboarding time by over 60%.

### Client Facing Software Engineer, Remotebase (Jul 2023 to Jan 2024)
Worked directly with product managers and clients to translate business requirements into working software on the Snapdev platform (now Amber.ai).

### Full Stack Engineer, Thathal I&TS (Sep 2023 to Aug 2024)
Built Google Ads API connectors and React/TypeScript dashboards. Co-developed a microservices-based enterprise SaaS platform, cutting deployment time by 40%.

## Education

Bachelor of Software Engineering, FAST-NUCES (2020 – 2024)

## Certifications

AWS Certified Cloud Practitioner