aj
Contact
Menu

V Ashish Joseph · GTM EngineerAzuga (a Bridgestone company)

I build and study decision systems.

Internal apps, reporting and AI review for go-to-market teams. Independent research into AI and data.

See the work Email Ashish
Three sources. One next step.EXAMPLE
  1. CRM recordThe account is evaluating.
  2. Meeting noteThe integration guide was requested.
  3. Email threadThe technical lead should be included.
Ready for reviewSend the guide. Include the technical lead.

The CRM record, meeting note and email thread supply the account context, guide request and technical lead. The draft remains subject to a person's review.

Sources stay attached to the next step
Explore 3D
01 / WORKSELECTED BUILDS & STUDIES

From an operational needto something people use.

Internal builds at Azuga, followed by independent research and design studies. Workflow sketches show each project’s purpose.

Azuga · InternalReleased to managers
Workflow sketch
  1. Support conversations
  2. Quality evaluation
  3. Manager review & coaching

Support quality & coaching intelligence

Helping support managers review conversation quality and identify what to coach.

From
Conversation analysis
To
Review criteria and coaching actions
MY PART

I built quality evaluation, parameter tracking and the manager review loop.

How the review works
  • Moved the output past a single quality score into tracked parameters a manager can actually coach against.
  • Designed the manager review loop first, so analysis quality could be tested before it reached frontline reps.
  • Kept a person at the end of it: the system proposes, the manager decides.

Manager feedback precedes wider rollout.

Email about this work
Azuga · InternalInternal build
Workflow sketch
  1. Reporting & workflows
  2. Role-based access
  3. One team interface

Internal GTM platform

One place for teams to use reporting and workflows, with access matched to their role.

From
Reporting and data access needs
To
A governed internal application
MY PART

I build the application, access controls and workflows teams use to reach their data.

Design decisions
  • Designed the application layer so teams reach what they need through a governed interface instead of direct database access.
  • Built read-only, table-scoped access with query visibility, so what gets exposed stays auditable.
  • Worked through authentication, role and ownership checks, and single sign-on.
Email about this work
independent researchWorking paper in progress
Workflow sketch
  1. Road crash records
  2. Spatial & sensitivity analysis
  3. Dashboard & policy briefs

Bengaluru road crash severity analysis

Comparing road crash severity across Bengaluru through spatial analysis and an interactive dashboard.

From
Road crash observations
To
Reproducible policy analysis
MY PART

I built the spatial analysis, interactive dashboard and reproducibility checks.

Methods and reproducibility
  • Built the spatial methodology — station-level geographies, Voronoi and H3 representations, and sensitivity analysis.
  • Wrapped it in a reproducibility framework: provenance tracking, a claim ledger, automated spatial checks and reproducible builds.
  • Wrote it for policy readers, with case studies and briefs rather than a dashboard on its own.

Analytical research, not an operational risk or policing system.

Email about this work
product architectureArchitecture study
Workflow sketch
  1. Spend & customer journeys
  2. Attribution comparison
  3. Architecture & MVP design

Marketing attribution & journey intelligence

Designing how to connect marketing spend, customer journeys and pipeline reporting.

From
Lifecycle and attribution questions
To
A proposed reporting architecture
MY PART

I modelled the B2B lifecycle and compared attribution approaches for the MVP design.

Attribution approach
  • Modelled the lifecycle from anonymous visitor through to customer, and connected spend to pipeline outcomes.
  • Compared first-touch, last-touch, linear and W-shaped attribution, with room for Markov and causal lift later.
  • Read the incumbents closely before designing anything, rather than inventing a model from scratch.

Architecture and MVP design work. Not a shipped product.

Email about this work

Internal work is shown through capability sketches.

02 / PLAYGROUNDTRY THE DECISION RULES

Two decisionsyou can inspect.

A score needs context. A draft needs evidence. These prepared examples show how I handle both.

ROUTING DEMO

Why is 96 held
while 78 is ready?

A high score cannot make an old record current. Change the threshold and inspect the reason behind each decision.

Test the review rules
Balanced rules · illustrative points
96HeldActivity is too old
78Ready for reviewRecent activity, category and fit

Fictional records from the routing example.

FOLLOW-UP DEMO

When context is missing, ask.

Remove the meeting note. The draft asks what to send instead of inventing a request. Every action stays tied to an available source.

Compare both versions

What changes with the meeting note

All three records

Current inputs
Integration guide
Prepare the integration guide.
Draft opening
“Following up on your request for the integration guide.”
Before sending
Attach the correct integration guide

Without the meeting note

Integration guide
Confirm the requested follow-up material.
Draft opening
“Following up on your evaluation. Could you confirm which information would help next?”
Before sending
Confirm the requested follow-up material

The technical lead stays included. The next-call date still needs confirmation.

01Inspect the inputsThree records. One conversation.

Each record contributes a different part of this follow-up. Drop the meeting note to see what happens when context goes missing.

CRM recordExample account · Evaluation · Sales Rep
Meeting note“Send the integration guide before the next call”
Email thread“Please include our technical lead.”
02Connect the recordsKeep each fact attached to its source.

The available records belong to one account. Follow a source link to inspect the record behind any fact.

Evaluation in progress
Source: CRM record
Integration guide requested
Source: Meeting note
Include the technical lead
Source: Email thread
03Draft a source-backed briefSeparate the next actions from the gaps.Integration guide included

Only supported actions belong in the brief. Missing information stays visible for the rep to resolve.

SOURCE-BACKED BRIEF

Prepare the integration guide.Source: Meeting note

Include the technical lead.Source: Email thread

Next-call date unknown. Ask the rep to confirm.

04Review the draftLeave the last decision with the rep.Integration guide included

The draft and checklist follow the available context. The rep still confirms the recipient, the gaps and the attachments.

DRAFT FOLLOW-UPNot sent

To: Example account team
Review: Confirm technical lead’s address

Hi team,

Following up on your request for the integration guide.

I’d like to include your technical lead in the conversation.

Could you confirm the best time for our next call?

Before sending
  • Attach the correct integration guide
  • Confirm the technical lead’s address
  • Confirm the next-call date
  • Review the draft before sending

Prepared scenario with sample records. Nothing is sent.

03 / RESEARCHEVALUATION, RELIABILITY, APPLIED AI

Independent research.A wider lens on decisions.

Published work spans text classification and computer vision. My current research focuses on AI evaluation and reliability.

Published papers

  1. Published

    Detecting cyberbullying in Twitter: a multi-model approach

    ICDECS · 2024

    A comparison of machine-learning models for classifying abusive behaviour in social media text.

    Read on ResearchGate Publisher record
  2. Published

    Precision Agriculture Meets AI: Utilizing Deep Learning Models for Accurate Tomato Leaf Disease Classification

    ICRISST · 2024

    A computer-vision study evaluating deep learning models for multiclass plant disease classification.

    Read on ResearchGate Publisher record
Current research · three studies in progress

Research in progress

  1. In progress

    Policy-sensitive AI decisionsPolicyBoundary-CS

    Testing whether AI decisions change for relevant policy facts and stay stable for irrelevant ones.

    Methods and status

    Human-gated by design: measurement scope, then hypotheses around evidence position, retrieval and service-preserving verification, then benchmark construction — each stage approved before the next. Model runs and human validation are not yet complete, and the write-up says so.

  2. In progress

    Auditing AI lead scoringLeadScoreAudit

    Auditing AI lead scoring for calibration, fairness and ranking robustness using public data.

    Methods and status

    Built on public data — UCI Bank Marketing plus synthetic B2B leads — specifically so practitioners can reproduce it without exposing anything of their own.

  3. In progress

    Reproducible AI evaluationRUNG evaluation

    Building reproducible AI evaluations through paired tests, independent audits and remediation.

    Methods and status

    An early audit returned a no-go. The package was remediated and re-audited under a separate focused review, with checksum validation, reviewer packets and a hard line between automated assessment and genuine human review.

ResearchGate profile Published papers on Google Scholar
04 / ABOUT

An engineer’s approach.A researcher’s questions.

I’m V Ashish Joseph, a GTM Engineer at Azuga (a Bridgestone company).

At Azuga I build internal tools, automations, dashboards and data flows for go-to-market teams.

The pattern is the same whether the subject is road crashes, support conversations, marketing spend or an AI benchmark: messy information becomes structured signals, signals become a decision, and a person still owns the last step.

Alongside that I run independent data and AI research, on evaluation, reliability and applied AI for revenue operations.

See the research
HOW I BUILD

Start with the decision. Keep the source records attached. Make the exceptions visible before asking anyone to act.

See it run in the playground
WHAT THAT LOOKS LIKE IN PRACTICE
01

Internal tools

Interfaces and workflows that bring what a team needs into one place, for the people actually doing the work.

  • Interfaces
  • Workflows
  • Revenue Operations
02

Connected data

Data flows and dashboards where the record behind an answer matters as much as the answer.

  • Data flows
  • Dashboards
  • Automation
03

Applied AI

AI placed inside a workflow with a human decision at the end, not bolted on as another window to chat with.

  • Applied AI
  • Evaluation
  • Human review