Talat Naz

Talat Naz

About

Detail

AI Engineer | LLMs | AI Agents | RAG | Python | FastAPI | LangGraph | LangChain | OpenAI | Azure AI | AWS | Production AI Systems | Open to remote contracts
Houston, Texas, United States
Contact Talat regarding: 
work
Full-time jobs

Timeline

work
Job
school
Education
folder
Project

Résumé

Jobs verified_user 0% verified
  • HurixDigital
    Agentic Ai Engineer
    HurixDigital
    Dec 2024 - Current (1 year 11 months)
    Architected and delivered enterprise Generative AI and Agentic AI solutions using LLMs, OpenAI, Azure OpenAI, Amazon Bedrock, RAG, LangGraph, LangChain, MCP, and vector databases.

    Designed multi-agent and tool-using AI systems with agent orchestration, tool calling, structured outputs, memory, retrieval, and enterprise API integrations.

    Built production-grade RAG applications and AI agents for enterprise use cases, optimizing retrieval, prompting, context management, response quality, latency, and cost.

    Led customer discovery, AI solution architecture, technical workshops, PoCs, architecture reviews, and executive presentations, translating business requirements into scalable GenAI and agentic solut
  • S
    AI Product Manager
    Supply Logistics Analysis System
  • Abbasai Solutions
    AI Engineer
    Abbasai Solutions
    Sep 2022 - Jun 2024 (1 year 10 months)
    Designed and deployed production Generative AI applications using Python, OpenAI, Anthropic, RAG, LangChain, LangGraph, Pinecone, and FAISS.

    Built AI agents and multi-step agentic workflows with LangGraph, tool calling, function calling, MCP, structured outputs, and dynamic workflow orchestration.

    Developed enterprise RAG pipelines covering document ingestion, chunking, embeddings, semantic retrieval, reranking, context optimization, and response generation.

    Built AI-powered copilots, chatbots, semantic search, document intelligence, and workflow automation solutions using LLMs and agentic architectures.

    Implemented LLM evaluation, guardrails, observability, caching, prompt engineering
  • VAZATA
    Artificial Intelligence Intern
    VAZATA
    Jan 2022 - Jun 2022 (6 months)
    Developed and supported AI and Generative AI applications using Python, OpenAI, Anthropic, LLMs, RAG, LangChain, LangGraph, and vector databases.

    Built RAG pipelines and AI-powered workflows using embeddings, semantic search, prompt engineering, and document processing to improve information retrieval and automation.

    Developed REST APIs and backend services with Python and FastAPI, integrating AI models and external services into application workflows.

    Experimented with AI agents, tool calling, function calling, and MCP to build intelligent workflows capable of interacting with tools, APIs, and enterprise data.

    Assisted with model evaluation, testing, guardrails, and observability to improve the a
Education verified_user 0% verified
  • University of Houston
    Bachelor's Degree, Computer Science
    University of Houston
    Sep 2018 - Sep 2022 (4 years 1 month)
Projects (professional or personal) verified_user 0% verified
  • F
    FinPulse — Multi-Tenant Financial Analytics API
    Aug 2023 - Current (3 years 3 months)
    Open-source backend that answers the question founders actually ask: "How many months of cash do we have left?" Pulls transactions from accounting providers (QuickBooks, Xero), normalizes them into a single ledger, and computes burn rate, runway, cash flow, and spend anomalies.
    ▸ Multi-tenant by construction — tenancy enforced in one dependency; cross-tenant probes return 404 (not 403) so organization IDs never leak
    ▸ Role-based access control (owner / admin / member) with JWT auth
    ▸ Provider adapters behind a single protocol — adding a new accounting provider touches one file; deterministic sandbox mode demos with zero credentials
    ▸ Idempotent syncs — re-importing never duplicates a ledger
    ▸ Explainable anomaly det
  • A
    AgentMesh — Agentic RAG Framework (Zero-Dependency Core)
    Lightweight agentic RAG framework in pure Python — the core has zero runtime dependencies. Instead of stuffing retrieved chunks into the prompt up front, AgentMesh exposes retrieval as a tool: the model decides when to search, reformulates queries, runs multiple searches for multi-part questions, and answers with [chunk-id] citations pointing back to the exact source passage.
    ▸ Provider-agnostic by design — everything model-specific hides behind a one-method LLMProvider protocol; Claude support ships as an optional extra, and swapping models means implementing one method
    ▸ Fully testable offline — a scripted provider replays deterministic turns, so the agent loop, tool dispatch, error recovery, and citation flow run as fast unit t
  • K
    KPI Canvas
    Financial KPI dashboard in React + TypeScript where every chart is a hand-rolled SVG component — no chart library. Built to demonstrate the layer beneath charting libraries: mark geometry, axis design, color systems, and interaction, all from a written spec.
    ▸ KPI stat tiles (cash on hand, monthly net with delta coloring, average burn, runway), a 12-month income-vs-expenses grouped bar chart with crosshair tooltips, and sorted category spend bars with direct labels
    ▸ Theme-aware validated palette — light and dark are separately chosen steps of the same hues, checked for colorblind-safe separation (not a CSS invert); every token swaps in one CSS block
    ▸ Accessible — charts carry role="img" with descriptive labels; text never w