Jay Dalvi
AI / Agentic WorkflowsInteractive Prototype

From fragmented relationship data to actionable network intelligence.

A lightweight product prototype exploring how AI-assisted workflows can turn opportunity requirements, network relationships, and engagement context into explainable recommendations and next actions.

Try the prototypeBack to selected work

This prototype is an anonymized, synthetic reconstruction created for portfolio purposes. It demonstrates the product concepts and workflow thinking without exposing proprietary data or implementation details.

The problem

Networks contain intelligence. Most tools return a list.

Professional networks contain valuable information about skills, relationships, past engagement, and expertise — but that context is often fragmented across tools and difficult to act on.

When a new opportunity appears, users need to quickly answer:

  • Who is relevant?
  • Why are they relevant?
  • Who has the strongest relationship?
  • What should we do next?

Traditional search surfaces people. The product opportunity is to surface actionable relationship intelligence.

Prototype

Network Intelligence Agent

Edit the opportunity, analyze, explore the network graph, review evidence, shortlist candidates, and draft outreach — all client-side with synthetic data.

Interactive prototypeSynthetic people and interactions — no real data.
Step 01 · Opportunity

Intake & requirements

Editable
Required
Preferred
Step 03 · Network

Relationship map

StrongMediumWeak
You

Click a node to see relationship path and evidence. Dashed edges are second-degree connections.

Activity

Audit trail

No activity yet. Analyze an opportunity to begin the trail.

How it works

A workflow, not an agent free-for-all.

Each step is labeled by its role: AI-assisted interpretation and explanation, deterministic logic for search and ranking, and human decisions at the moments that matter.

AI-assistedRules / structured logicHuman decision
  1. 01Opportunity
  2. 02Interpret
  3. 03Search network
  4. 04Rank relevance
  5. 05Explain
  6. 06Human review
  7. 07Next action
Key product decisions

Four decisions that shaped the surface.

  1. 01

    Rank with evidence, not just a score.

    Users need to understand why someone is being recommended, not just be handed a number they can't audit.

  2. 02

    Relationships matter alongside capability fit.

    The best candidate is not always the most actionable connection — the shortlist should weigh both.

  3. 03

    Keep humans in control of outreach.

    Relationship actions carry reputational context that should not be automated blindly.

  4. 04

    Make the network path visible.

    Understanding who can make the strongest introduction is often more valuable than finding another name.

Risks & controls

Where automation should stop.

Potential risks

  • Incorrect relationship inference
  • Stale interaction data
  • Overweighting network strength
  • Missing relevant candidates
  • Sensitive communication context
  • Automation acting without approval

Controls

  • Human review before outreach
  • Visible evidence for every recommendation
  • Synthetic / structured reasons instead of opaque scores
  • Manual override on every step
  • Activity logging and audit trail
  • No autonomous sending
  • Clear source and context labeling
Metrics I would track

What good would look like.

Proposed product metrics — not claimed historical outcomes. The goal is a signal loop that surfaces where the product creates leverage and where humans still have to work around it.

  • Time to useful shortlist
  • Recommendation acceptance rate
  • Shortlist-to-introduction rate
  • User override / dismissal rate
  • Percentage of recommendations with usable evidence
  • Time from opportunity intake to outreach
  • Repeat usage
Rollout

Increase autonomy only after users trust the recommendations.

  1. Phase 1

    Recommendation assistant

    • — Opportunity interpretation
    • — Network search
    • — Evidence
    • — Human shortlist
  2. Phase 2

    Action assistance

    • — Warm-intro suggestions
    • — Draft outreach
    • — Relationship-owner routing
  3. Phase 3

    Workflow automation

    • — Follow-up reminders
    • — Task creation
    • — Approved integrations
My contribution

Where I was involved.

  • Shaped workflow concepts around skills and agents
  • Worked on agency / opportunity intake
  • Explored network and engagement signals
  • Worked around relationship intelligence concepts
  • Contributed to Gmail integration–related product thinking
  • Translated ambiguous workflow ideas into product requirements and user experiences

Cross-functional collaboration

  • Product
  • Design
  • Engineering
  • Stakeholders
Back to selected workHomepage