This year i.e. 2026, here, I am building a portfolio of agents.
Portfolio of agents of Hari Om Vashishtha
Agent 01
God Complexes — Hiring Alignment System
ConceptSince I needed businesses on top of which I could build agents, for my first ever agent, I picked up the universal problem I have faced throughout my career — Hiring Alignment.
It’s a half-baked 55 page thesis that proposed a new way how the hiring industry could evolve in the age of information being omnipresent. That felt like a far-fetched dream project so I came down to the ground to learn to build daily use agents.
Agent 02
Live · Awareness stage · v1For the second agent, I used my long-term eternal dream project — building a discipline programme for kids — as the foundation of my Salesforce agents. The programme (3MistakesOfMyLife.in) is designed to develop structured thinking, self-awareness, and deliberate practice through guided conversation rather than lectures. Parents and mentors participate alongside kids. Discipline becomes a measurable practice, not a vague concept.
The agent is the conversational front door to the system. It currently handles one stage: awareness. One job done properly before expanding to the next.
Architecture — current state
- RouterTopic Selector — reads user intent, routes to the correct conversation path
- RAGProject FAQs — retrieves answers from a structured knowledge library grounded in actual project documentation
- BoundaryOff-topic handler — gently redirects out-of-scope questions, protecting system integrity
Layer separation principle
Layers are deliberately separate. The conversation interface can evolve without breaking the data structure underneath.
Agent 03
Hermes — Million Dollar Consistency Partner
v3 · Active build · Daily operational useThis is where it all started. A custom GPT (v0.1) built to keep me aligned to the year-end goal — but it couldn’t hold the complexity of a daily build practice. So I rebuilt it as a proper multi-agent pipeline.
Hermes runs after every live YouTube build session. It pulls the transcript autonomously, processes it through five specialised agents in sequence, and pushes a fully structured debrief to Notion — without me touching documentation. The focus stays on building, not on capturing what was built.
5-agent pipeline
- Agent 1Parser — extracts explicit tasks, implicit tasks, artifacts, learning, ICP interactions, and content created from the raw transcript
- Agent 2Alignment engine — classifies every activity into Noise / Learning / WIP / Artifact / Revenue signal. Scores the day. Verdict: aligned, weak, or misaligned
- Agent 3Summary generator — plain-language debrief: what was attempted, what was produced, what moved toward revenue, what was wasted motion
- Agent 4Task reconciler — merges carry-forward tasks from the previous session with today’s output. Every task categorised as Leverage, Overhead, or Neutral
- Agent 5Next actions generator — recommends 3–5 highest-leverage actions for the next session. Each requires a traceable revenue connection. No hedging
Infrastructure
Now the question is, how many useful ones can I build by the end of 2026??? 🙂
Agents Are Coming, Just Blend In Before It’s Too Late