I build practical AI systems that can be explained to operators, shipped by teams, and trusted with real tools.
I bridge the technical and non-technical sides of AI work: product strategy, model testing, research workflows, internal automation, community, and hands-on engineering. My current lane is agent tooling, MCP security, OpenClaw/Pokee skills, procedural memory, and SEA-focused automation products.
- Pokee/OpenClaw skills: building practical agent skills and integrations that connect deep research, product workflows, and real users.
- Model testing and research ops: working across model behavior, researcher workflows, internal processes, and usable agent experiences.
- MCP security and reliability: making agent tool calls observable, rate-limited, auditable, and safer by default.
- Agent memory: exploring Hermes-style procedural memory so agents can reuse workflows instead of relearning them every session.
- SEA automation products: turning local operational pain into small, shippable software and service offers.
OpenClaw skill for Pokee AI's Deep Research API.
At Pokee AI, I work across product, model testing, research workflows, internal workflows, and agent integrations. This skill is one public artifact from that lane: making Pokee's deep research agent usable inside OpenClaw workflows for strategy, market research, and due diligence.
The broader work is about bridging researchers, technical builders, and non-technical users: testing model behavior, improving internal processes, translating research capabilities into usable agent experiences, and building tools that help people actually use advanced agents.
Links:
Local-first firewall and black-box recorder for MCP-connected agents.
- Blocks poisoned MCP tool descriptions before agents can see them
- Detects descriptor drift after a trusted baseline
- Rate-limits runaway tool calls
- Holds risky actions for approval
- Produces redacted audit logs and share-safe run reports
This is the core thesis I am building around: production agents need infrastructure around tool access, not only better prompts.
Procedural memory for AI agents.
Myelin observes repeated agent workflows, clusters action sequences, promotes reusable procedures, and exposes learned behavior through MCP. It is built for agents that should get better at the work they actually do.
Philippines-focused BidOps monitor for public procurement opportunities.
It started as a scraper and evolved into a commercial product angle: help suppliers decide which government bids are worth pursuing, reviewing, or skipping before wasting time on weak-fit opportunities.
Philippines-focused Meta ads manager prototype.
AdPulse PH explores a local-market SaaS angle for advertisers, media buyers, and agencies managing Facebook ad accounts, pages, exports, and dashboard workflows.
Predecessor to Myelin.
Sigil explored local-first agent memory, SQLite-backed knowledge graphs, personas, orchestration, and multi-agent coordination. Myelin is the sharper successor focused on procedural memory.
I am not only interested in technical demos. I care about adoption, positioning, and the path from prototype to operational value.
- Product/GTM and AI workflow work with Pokee AI.
- AI strategy and operations work across portfolio workflows, automation standards, and internal AI rollout.
- Salesforce AgentForce and workflow automation experience across CRM, support, SOPs, dashboards, and low-code systems.
- Community, onboarding, and launch work that turns AI capabilities into something users can understand and adopt.
Technical: Python, TypeScript, MCP, local-first tooling, SQLite, RAG, LLM workflows, agent orchestration, automation pipelines, security-minded tool design.
Product: AI product strategy, roadmapping, user onboarding, workflow mapping, GTM experiments, community feedback loops, founder/SMB automation offers.
Operations: turning ambiguous processes into tools, dashboards, SOPs, training materials, and repeatable execution systems.
This account is a build trail. Some repos are polished products, some are research prototypes, and some are product experiments. The connective tissue is practical AI:
- Can an agent use this safely?
- Can a human operator understand it?
- Can a team adopt it without magic?
- Can it become public proof, a product, or a client offer?
- Portfolio: niamportfolio.netlify.app
- LinkedIn: linkedin.com/in/niam-amor