Junior AI Engineer | LLM applications, RAG, agents and voice AI
Computer Science student at FAST NUCES Islamabad. I build AI systems end to end: the models, the backends, the evaluation and the interface around them.
I work as a Junior AI Engineer and study Computer Science at FAST NUCES. I like LLM systems that are measured, guarded and built to keep running, not just demos. Here is what I have learned by building them:
- Real-time voice agents: streaming speech-to-text, LLM and text-to-speech pipelines over WebSockets, with barge-in handling and low latency (Deepgram, OpenAI, ElevenLabs).
- LLM backends: async FastAPI services with tool calling, conversation memory, Valkey caching and rate limiting, server-sent events and background jobs.
- Retrieval-augmented generation: corrective RAG over long SEC filings with pgvector and LangGraph, cited answers, abstention when the evidence is missing, and accuracy tracked on a hand-built evaluation set.
- MCP servers and guardrails: tool servers for calendars and for inventory and orders, with NeMo Guardrails on input and output, tested against jailbreak and prompt-injection attempts.
- Agents: multi-agent pipelines in LangGraph, such as DevMind, which turns a GitHub issue into a tested pull request.
- Also: computer vision and automation (OpenCV, MediaPipe, OCR, Playwright), and fundamentals built from scratch in C++, Java, Python and x86 assembly.
Currently exploring system design for MCP and agent-to-agent (A2A) architectures.
| Area | Technologies |
|---|---|
| Languages | Python, JavaScript, Java, C++, SQL, x86 Assembly |
| AI and LLM | LangChain, LangGraph, RAG, MCP, NeMo Guardrails, tool calling, multi-agent systems |
| Voice and backend | FastAPI, Flask, WebSockets, SSE, Deepgram, ElevenLabs, Docker |
| Data | PostgreSQL, pgvector, ChromaDB, Valkey, SQLite |
| Vision and automation | OpenCV, MediaPipe, PyTorch, EasyOCR, Playwright |
| Tools | Git, GitHub, pytest, Postman, Streamlit |