R
Most PMs run programs. I also build the tools the team uses to run them.
Five years running data and AI programs end to end. Six AI agents in production, adopted across the full team in under a month.
Most PMs run programs. I also build the tools the team uses to run them.
At phData I run data and AI programs end to end — not as a coordinator, but as the person accountable for scope, budget, risk, and the client relationship when things go sideways. I've managed the stakeholders most likely to derail an engagement, and turned completed programs into renewals.
I also built six AI agents now running across the delivery team. Billing, sprint health, risk tracking, executive decks — automated so the team's time goes to the work that actually moves programs forward.
Core Competencies
Five years leading cross-functional teams through chaos. Three years building AI agents that automate the work no one should be doing manually.
Program Manager
Cross-Functional Leadership
Engineering, finance, product, client — all moving in the same direction
Blocker Removal
Find what's slowing teams down and clear it before it shows as a delay
Cross-Team Communication
Translate between teams who don't naturally speak the same language
Delivery Awareness
Know where every track actually stands, not just what's been reported up
Navigating Chaos
When things go sideways: diagnose fast, decide, keep the team moving
Escalation Path Design
Right person called at the right time. Severity classified before it gets raised.
Change Impact Assessment
Downstream effects across all tracks evaluated before any change gets signed off
Agentic AI Specialist
Claude Agent Building
Built billing, sprint health, risk, status, and budget agents for PM teams
Weekly Task Automation
Automated 4–6 hrs/week of manual PM work per person via scheduled agents
MCP Tool Integration
Jira · Salesforce · Glean · Google Workspace connected and running
AI Adoption
0 to 14 PM adopters. Cohort rollout, 1:1 onboarding, usage tracked per person.
AI Skilling
Taught team to use and build agents; two built independently from my docs
Agent Output Review Design
Human review checkpoints built in. Wrong outputs caught before they reach anyone.
PSM I
Scrum.org
ITIL 4
Axelos
5+ yrs
phData
4
programs
Two roles, one throughline: the work that keeps a program on track is rarely the work that's written down. Process governance at BT, full program ownership at phData — different domains, same instinct for spotting what's about to go wrong before it does.
Education
Bachelor of Computer Application
Brainware University, Kolkata
Award
phData Innovation Award
Recognised for building the delivery team's AI agent platform — Glean + Claude + n8n agents handling month-end billing reconciliation, sprint health scoring, project health scans, and leadership deck drafting. Recovers about 8 hours a week across the team.
Certifications
PSM1 — Professional Scrum Master
Scrum.org
ITIL 4 Foundation
Axelos
Building Systems with the ChatGPT API
DeepLearning.AI
LandingLens Computer Vision Fundamentals
LandingAI
DEC 2022 — PRESENT
CURRENTphData · Snowflake Elite Partner · Kolkata
AUG 2020 — DEC 2022
British Telecom (BT) · Kolkata
Education
Bachelor of Computer Application
Brainware University, Kolkata
2017 – 2020 · CGPA: 7.5 / 10
Certifications
PSM1 — Professional Scrum Master
Scrum.org
ITIL 4 Foundation
Axelos
Building Systems with the ChatGPT API
DeepLearning.AI
LandingLens Computer Vision Fundamentals
LandingAI
Briefing to Big Data & Hadoop Ecosystem
Coursera
Programming in Java
NPTEL
Award
phData Innovation Award
Recognised for building the delivery team's AI agent platform — Glean + Claude + n8n agents handling month-end billing reconciliation, sprint health scoring, project health scans, and leadership deck drafting. Recovers about 8 hours a week across the team.
Most of what makes a program succeed doesn't show up as a number — it's a conversation that happened a week earlier than it had to, or a cadence redesigned around a problem instead of fought. These four are the closest thing to a paper trail for that kind of work.
01
0–5Programs running at the same time
Each at a different stage — one in kickoff, one mid-cadence, one heading into closure — switching context fast enough that none of them feel like the neglected one.
02
0.0 HRSTimezone gap, US to China, one program
A US delivery lead to a client's factory site in China, on a single program — redesigned the whole cadence around the gap instead of fighting it.
03
0%Of the PM team using the agents within one month
Every PM on the team was using it within a month of shipping — because it solved their problem too, not just mine.
04
0Programs, three different definitions of done
Fixed-price delivery, long-term retainer, and an internal build — each needs a different conversation about risk, budget, and what 'green' actually means.
01
0–5Programs running at the same time
Each at a different stage — one in kickoff, one mid-cadence, one heading into closure — switching context fast enough that none of them feel like the neglected one.
02
0.0 HRSTimezone gap, US to China, one program
A US delivery lead to a client's factory site in China, on a single program — redesigned the whole cadence around the gap instead of fighting it.
03
0%Of the PM team using the agents within one month
Every PM on the team was using it within a month of shipping — because it solved their problem too, not just mine.
04
0Programs, three different definitions of done
Fixed-price delivery, long-term retainer, and an internal build — each needs a different conversation about risk, budget, and what 'green' actually means.
TEAM-WIDE AI AUTOMATION
Before
About 8 hours per week of PM time across active programs went into the repeatable parts of the job — month-end billing reconciliation, sprint health scoring, project health scans, leadership deck drafting — done manually, program by program.
After
6 agents I designed and built now run across the delivery team on Glean, Claude, and n8n — recovering ~8 hours per week, with the billing agent accurate above 95% on its discrepancy check. Built because I wanted those hours back. Adopted because the rest of the team wanted theirs back too.
04 / WORK
Four programs. Each card is the real story — the non-obvious thing that made each one hard, and what I did about it.
05 / AI AGENTS
Month-end billing reconciliation was 3–4 hours of copy-paste per PM, per program — no judgment required, just a machine that hadn't been written yet. Built with accuracy checks and human-review fallbacks before it touched a live client invoice. Cut billing time by 60%, accurate above 95%, adopted team-wide inside a month. Five more agents followed. Running programs while building the tooling means every agent gets built for the real problem — not the documented version of it.
60% time saved · >95% accuracy · adopted team-wide in under a month
CSV → ranked scorecard + PM action items in under 2 min
Replaces a 2-hour manual checklist across 4+ active programs
Executive deck drafted in under 20 min · CFO vs VP Engineering tone
60% time saved · >95% accuracy · adopted team-wide in under a month
CSV → ranked scorecard + PM action items in under 2 min
Replaces a 2-hour manual checklist across 4+ active programs
Executive deck drafted in under 20 min · CFO vs VP Engineering tone
The story behind it
Why I built it
End-of-month billing reconciliation was 3-4 hours of copy-paste between the time-tracking tool and the project budget, every program, every month. I'd done it by hand for two years. The data was structured and the process was identical every time — that's not PM work, that's a script that hasn't been written yet.
Build vs. buy
Built on Glean, not a custom API integration — it was already the team's knowledge and permissions layer, so an agent there could see live program data, Salesforce, and Drive without a separate integration build. Less impressive on paper, far faster to ship and far easier for the rest of the team to pick up.
Adoption sequencing
Shipped the billing agent first because it was the highest-pain, most-measurable win — if it didn't hold up, nothing after it would get adopted either. Once it was running team-wide inside a month, the next five had an easier path in.
Why the accuracy number is real
The >95% figure on the billing agent isn't a vibe — it's backed by per-run accuracy logs checked against a test set of past reconciliations, the same way I'd back up any number before it went into a status report to a client.
Why I built it
End-of-month billing reconciliation was 3-4 hours of copy-paste between the time-tracking tool and the project budget, every program, every month. I'd done it by hand for two years. The data was structured and the process was identical every time — that's not PM work, that's a script that hasn't been written yet.
Build vs. buy
Built on Glean, not a custom API integration — it was already the team's knowledge and permissions layer, so an agent there could see live program data, Salesforce, and Drive without a separate integration build. Less impressive on paper, far faster to ship and far easier for the rest of the team to pick up.
Adoption sequencing
Shipped the billing agent first because it was the highest-pain, most-measurable win — if it didn't hold up, nothing after it would get adopted either. Once it was running team-wide inside a month, the next five had an easier path in.
Why the accuracy number is real
The >95% figure on the billing agent isn't a vibe — it's backed by per-run accuracy logs checked against a test set of past reconciliations, the same way I'd back up any number before it went into a status report to a client.
Animated flow diagrams · Architecture · Tech stack
I'm a Program Manager, not a hands-on engineer. What follows is the technology surface area I've owned delivery across — enough to scope, sequence, and govern risk without needing the engineering team to translate everything.
The engineering belongs to the team. The plan, the controls, and the outcome belong to me.
Data platforms delivered on
Engineering tooling in-program
Delivery tools I run daily
AI & automation layer I've built
LiveThis stack powers the 6 AI agents currently running in production across the phData PMO.
Data platforms delivered on
Engineering tooling in-program
Delivery tools I run daily
AI & automation layer I've built
Live in productionThis stack powers the 6 AI agents currently running in production across the phData PMO.
Read this as delivery surface area, not a technical skills inventory. My value is in program design, financial governance, and risk anticipation across these technologies — not in writing the code that runs on them.
07 / SOLUTION DESIGN
The six agents above are already running. This shows how I approach a new one — from the business problem to the data model to the rollout sequence — before a single line of automation is written.
A design walkthrough, not a delivered program — the PMO agent platform (above) is where the delivered numbers live. This shows the architecture and rollout thinking behind a new agent, end to end.
Platform build, GenAI delivery, or an engagement that needs picking up mid-stream — happy to talk through what the work actually is before anything else.
OPEN FOR PROGRAMS
Kolkata, India · IST · Available globally
Platform Build or Migration
Snowflake, Databricks, AWS. I own the program end to end — kickoff through closure, covering the weekly delivery rhythm, risk management, stakeholder communication, and financials.
GenAI Program Delivery
LLM integrations and AI agent builds. Structured delivery, real governance, and the same stakeholder accountability as any other program — because "it's AI" isn't a reason to skip the controls.
Mid-Stream Pickup
Programs where the previous PM is stepping off. I get current on scope, team, risk, and financials fast, and take the program from there without disrupting delivery.
Platform Build or Migration
Snowflake, Databricks, AWS. I own the program end to end — kickoff through closure, covering the weekly delivery rhythm, risk management, stakeholder communication, and financials.
GenAI Program Delivery
LLM integrations and AI agent builds. Structured delivery, real governance, and the same stakeholder accountability as any other program — because "it's AI" isn't a reason to skip the controls.
Mid-Stream Pickup
Programs where the previous PM is stepping off. I get current on scope, team, risk, and financials fast, and take the program from there without disrupting delivery.