Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
February 07, 2026
·
Dhaka
Building a Persistent Memory & Stateful Second Brain AI Agent
Learn how to build a stateful AI agent with persistent memory using Obsidian and context engineering. This talk demonstrates practical techniques for agent behavior and tool use without complex infrastructure.
Overview
Demonstrating context engineering in practice through Slatekore - an open-source starter kit that gives Gemini CLI persistent memory using Obsidian as the storage layer.
- State Management Without Infrastructure How to use the file system as persistent state - your Obsidian vault becomes both the agent’s memory and its knowledge base. No databases, no vector stores, no MLOps.
- Context Engineering for Agent Behavior Crafting system prompts (
GEMINI.md) and workflow files that define agent capabilities, constraints, and multi-step reasoning patterns. The prompt is the agent’s configuration.
- Tool Use Through Natural Interfaces Connecting the agent to real actions: creating files, managing tasks, building knowledge graphs - all through natural language commands backed by structured templates.
Watch a stateful agent capture research, recall context from previous sessions, update its knowledge graph, and execute project workflows - without any model training or fine-tuning.
Links
Slatekore automates Obsidian research workflows using Gemini CLI agent configurations.
Slatekore automates Obsidian research workflows using Gemini CLI agent configurations.
Tech stack
- workflow filesYAML-based configuration files that automate software development lifecycles by defining specific triggers, jobs, and execution environments.Workflow files function as the declarative blueprint for CI/CD pipelines (typically stored in the .github/workflows directory). These YAML documents orchestrate complex sequences: triggering on push or pull_request events, spinning up virtual runners (like ubuntu-latest), and executing discrete steps such as npm install or docker build. By version-controlling these configurations alongside source code, teams ensure every commit validates against specific linting rules and deployment targets. It is the industry standard for modern automation: defining exactly how code moves from a local branch to a production environment.
- LLVMLLVM is a modular compiler and toolchain infrastructure, providing reusable, language-agnostic components for high-performance compilation and optimization.LLVM is a collection of modular compiler and toolchain technologies, designed for maximum reusability and performance. Its core is the Intermediate Representation (IR), a portable, high-level assembly that acts as the optimization target between the frontend and backend stages. This three-part architecture allows projects like Clang (for C/C++) and compilers for Swift and Rust to share a single, powerful optimization and code generation engine. This flexibility enables efficient, cross-platform compilation for diverse architectures: from x86 and ARM to GPUs, making LLVM a foundational component in modern operating systems and development toolchains.
- Gemini CLIGemini CLI is your open-source AI agent: it brings the power of Gemini models directly into your terminal.The Gemini CLI is an open-source, terminal-first AI agent providing direct, lightweight access to the Gemini model capabilities. It operates as an interactive Read-Eval-Print Loop (REPL), enabling developers to execute complex tasks like code generation, project analysis, and bug fixing through natural language prompts. The CLI leverages a Reason and Act (ReAct) loop, orchestrating built-in tools (file system operations, shell commands, web search) and custom extensions via the Model Context Protocol (MCP). Users can access powerful models, including Gemini 2.5 Pro and Gemini 3 Pro, for high-performance agentic coding and streamlined workflows, all managed from the command line.
- AI agentAI agents are autonomous software systems (driven by LLMs) that plan, reason, and execute complex, multi-step tasks without continuous human oversight.This technology represents the next major leap in AI: autonomous execution. Unlike basic chatbots, agents use a Perceive-Decide-Act-Learn cycle to achieve high-level goals. They integrate tools (APIs, databases, web search) to perform complex workflows like IT automation, software design, and end-to-end lead generation. Platforms like AutoGPT and LangChain enable this development, moving AI from an assistant role to a proactive, digital worker. The market reflects this impact: it hit $7.6 billion in 2025 and is projected for 49.6% annual growth through 2033.
- GeminiGoogle's natively multimodal AI model: understands and operates across text, code, audio, image, and video.Gemini is Google's most capable and general AI model, engineered from the ground up to be natively multimodal: it seamlessly understands and combines information across text, code, audio, image, and video inputs. The technology is optimized for flexibility, running efficiently on everything from data centers to mobile devices. It is deployed in three key sizes: Ultra (for highly complex tasks), Pro (for broad scaling), and Nano (for efficient on-device tasks). Developers access this power via the Gemini API to build next-generation applications.
- System promptThe System prompt is the foundational instruction set: it defines the Large Language Model's (LLM) persistent role, behavior, and constraints.System prompts are the AI's 'job description,' operating behind the scenes to dictate model conduct. They establish the core persona, like an 'expert in Node.js and Express' or a 'customer interview coach.' This static instruction set ensures consistency across all user interactions, unlike dynamic user prompts. Effective system prompts integrate key elements: Role Definition, Behavioral Guidelines, and Ethical Constraints. For example, a prompt might mandate a 'professional and friendly tone' while setting a hard rule: 'Never share sensitive data.' This structure is critical for production-grade agents (ChatGPT, Claude, etc.), directly impacting reliability, safety, and adherence to a specific task domain.
- System promptsSystem prompts are the foundational, persistent instructions that define an LLM's persona, constraints, and overall behavioral alignment.System prompts function as the 'BIOS' configuration for a Large Language Model (LLM): they are a foundational instruction set, loaded once, that dictates the model's core conduct (e.g., 'You are a concise, trustworthy summarization engine'). Unlike a user prompt, the system prompt remains constant across all interactions, ensuring consistency. Developers use this mechanism to set specific guardrails, assign a role (e.g., 'expert TypeScript engineer'), and enforce output formatting (e.g., 'always use Markdown for code blocks'). This steerability is critical for production-grade AI applications, directly impacting response reliability and brand voice adherence.
- ObsidianObsidian is a powerful, local-first knowledge base: it uses plain Markdown files for notes and links them bi-directionally to build a personal, interconnected knowledge graph.Obsidian is a cross-platform note-taking application designed for building a 'second brain.' It operates on a local folder of plain Markdown files (a 'vault'), ensuring you own your data and maintain long-term portability. The core strength lies in bi-directional linking, which connects concepts using the `[[link]]` syntax, allowing you to visualize relationships via an interactive Graph View. With thousands of community plugins and themes, users can customize the platform for diverse workflows, from Zettelkasten to project management, making it a flexible and future-proof tool for serious knowledge workers.
- MarkdownMarkdown is a lightweight markup language that converts plain text into structured HTML using simple punctuation.John Gruber released Markdown in 2004 to simplify web writing for non-coders and developers alike. It uses intuitive syntax (like # for headers or * for lists) to produce clean XHTML. This standard now powers documentation on GitHub, discussions on Reddit, and static sites via Jekyll. By prioritizing readability, Markdown ensures that a source document remains legible even before it hits a browser.
- PythonPython: The high-level, general-purpose language built for readability, powering everything from web backends to advanced machine learning models.Python is the high-level, general-purpose language prioritizing clear, readable syntax (via significant indentation), ensuring rapid development for any team . Its ecosystem is massive: use it for robust web development with frameworks like Django and Flask, or leverage its power in data science with libraries such as Pandas and NumPy . The Python Package Index (PyPI) provides thousands of community-contributed modules, offering immediate solutions for tasks from network programming to GUI creation . The language is actively maintained by the Python Software Foundation (PSF), with the stable release currently at Python 3.14.0 (as of November 2025) .
- Prompt EngineeringPrompt Engineering is the discipline of structuring inputs (prompts) to Large Language Models (LLMs) to reliably and efficiently elicit a desired, high-quality output.This is the core skill for maximizing performance from models like GPT-4 and Claude 3: it's the art and science of guiding an AI. The process involves systematic iteration and applying specific techniques to control the model's behavior and reduce 'hallucination.' Key advanced methods include Chain-of-Thought (CoT) prompting, which forces the LLM to process complex problems step-by-step, and Few-Shot prompting (providing 2-3 examples) to establish a clear output format or style. Mastery of these methods directly translates to tangible gains: improved accuracy, reduced API costs from fewer retries, and production-ready outputs for applications like customer service bots or code generation.
- GCCThe industry standard compiler suite for C, C++, and Fortran: essential for building the modern Linux stack.GCC (GNU Compiler Collection) is the backbone of the open source world: it compiles the Linux kernel and thousands of critical software packages. It supports a wide array of frontends (including C, C++, and Ada) and targets architectures from x86_64 to RISC-V. Developers rely on its mature optimization passes (-O2 and -O3) to squeeze performance out of hardware. Version 13.x maintains this lead by refining C++23 support and improving static analysis tools to catch memory leaks early.
- LLVM IRLLVM IR is a strictly typed, RISC-like intermediate language that serves as the universal bridge between high-level source code and machine-specific binary instructions.This Static Single Assignment (SSA) based representation acts as the core interface for the LLVM compiler infrastructure. It utilizes a language-independent instruction set (including operations like `phi`, `getelementptr`, and `icmp`) to facilitate sophisticated middle-end optimizations. By standardizing the communication between front-ends like Clang or Rust and back-ends for x86, ARM, or RISC-V, LLVM IR ensures that a single optimization pass benefits multiple architectures. The format exists in three interchangeable forms: an in-memory C++ object model, a compact bitcode file (.bc), and a human-readable assembly text (.ll).
- RISC-VRISC-V is the open-standard Instruction Set Architecture (ISA) that eliminates proprietary licensing fees to enable custom silicon innovation.Managed by RISC-V International (a Swiss-based non-profit), this modular architecture allows developers to build everything from low-power microcontrollers to high-performance data center processors. It supports 32, 64, and 128-bit address spaces. Industry giants like NVIDIA and Western Digital already ship millions of RISC-V cores annually. The design relies on a small base ISA (fewer than 50 instructions) supplemented by specialized extensions: such as 'V' for vector processing or 'C' for compressed instructions. This flexibility lets engineers optimize hardware for specific workloads (AI, IoT, or automotive) without the restrictive costs or vendor lock-in associated with ARM or x86.
- x86_64A 64-bit instruction set architecture (ISA) that expands x86 registers and memory addressing while maintaining full backward compatibility.Originally launched by AMD as AMD64 in 2003, x86_64 serves as the standard 64-bit instruction set for modern desktop and server CPUs. It expands the legacy 32-bit x86 architecture by doubling the number of general-purpose registers to 16 and increasing their width to 64 bits (using prefixes like RAX and RDI). This evolution removes the 4 GB memory ceiling of 32-bit systems: it supports a theoretical limit of 16 exabytes of RAM. Most implementations utilize a 48-bit or 52-bit virtual address space to handle current enterprise workloads. The architecture remains the backbone of global computing by maintaining native compatibility with legacy 16-bit and 32-bit software through its versatile Long Mode execution state.
- ARMARM designs the high-efficiency RISC architectures powering everything from Apple's M-series silicon to billions of IoT sensors.ARM provides the blueprint for modern computing: a Reduced Instruction Set Computer (RISC) architecture optimized for performance-per-watt. While x86 handles legacy desktops, ARM defines the mobile and embedded landscape (powering 99% of the world's smartphones). Their licensing model allows partners like Qualcomm, Nvidia, and Amazon (AWS Graviton) to build custom silicon for specific workloads. With over 280 billion chips shipped to date, ARM is the industry standard for energy-efficient processing across data centers, automotive systems, and edge devices.
- AVX2A 256-bit SIMD instruction set expansion that doubles integer throughput and introduces FMA3 for accelerated vector processing.AVX2 (Advanced Vector Extensions 2) launched with Intel's Haswell architecture in 2013 to expand 128-bit integer SIMD instructions to full 256-bit widths. It adds critical features: Fused Multiply-Add (FMA3) for high-precision math and gather support for fetching non-adjacent data from memory. This instruction set is the standard for performance-critical software: 4K video encoding (x265), scientific simulations, and cryptographic hashing. By handling 32 bytes of data per instruction, AVX2 provides a significant efficiency ceiling over legacy SSE4.2 implementations.
- SSESSE (Server-Sent Events) enables real-time, uni-directional data streaming from servers to browsers over standard HTTP.SSE (Server-Sent Events) utilizes the EventSource API (standardized in HTML5) to establish a persistent, low-latency connection for server-to-client updates. It operates over standard HTTP/1.1 or HTTP/2: this bypasses complex firewall configurations and eliminates the overhead of WebSockets for one-way feeds. Engineers deploy it for high-frequency data (like live stock tickers or social media notifications) by setting the Content-Type header to text/event-stream. The protocol handles automatic reconnection natively: it uses structured fields (including data:, id:, and retry:) to ensure consistent delivery without manual polling or custom retry logic.
Finding related talks...
Compose Email
Sending...
Email preview
Loading recent emails...