Kumar Aniket · Senior Product Manager · Bengaluru
I ship AI-native
enterprise products.
Making AI spend legible.
Always on the hardest data problems in the room — every demo on this site is a product I shipped. Currently building FinOps for AI at Amnic.
BlockDeliver — decentralized CDN marketplace
Products I've shipped run inside teams at
Some things I've built
Show, don't tell.
Each build is a claim about what an AI product manager should know — LLM economics, data LLMs can use, agentic workflows — shown as a working, simulated slice of a real product. The numbers are illustrative, the product thinking is not.
Every AI feature has a meter running. Most teams can't read theirs — so I built one that makes spend legible, token by token.
AI Token Management
Enterprise LLM spend, made visible and governable — one view across every major model provider, with real-time anomaly detection, unit economics and cost attribution down to teams and features.
BedrockUsed by 10+ enterprises
Read the case study →AI spend · this month
$31.4k↑ 12% vs last month· 102M tokens
Anomaly flagged in real time +212%
Gemini 2.5 Pro · Growth team · Summarization feature
Likely cause: retry loop after provider timeouts — prompt tokens 4× baseline
attribution across models, teams, features & unit economics · illustrative
A model is only as smart as what it can reach. The quiet craft is turning messy customer feedback into something an LLM can actually query.
Customer Knowledge Graph
Every piece of feedback linked to the user, account and opportunity behind it — with source fields like ARR, plan type and deal stage kept searchable. A continuously-learning taxonomy classifies it all, so “what are customers saying?” becomes “what is it costing us?”. No manual tagging.
50+ enterprise customers
Used by teams at
“Exports keep timing out”
Collect
feedback pours in from 50+ sources
Connect
linked to the user, account & revenue behind it
Classify
auto-tagged by a taxonomy that keeps learning
Inside the taxonomy · a music app's feedback, illustrative
five tiers narrow 24.5k records to the exact topic
Invite links expire too quickly
Add QR-code invites
“Editing together just works”
themes carry sentiment — complaint, ask, praise
Agents aren't magic — they're workflows with judgment. The real design question is where the human stays in the loop.
FinOps Agents
Ask what's going on with your RDS bill in plain English — the agent walks the same modules a human FinOps analyst would: pull the costs, scan for anomalies, find the savings, draft the dashboard. A human approves before anything ships. I prototyped it in Claude Code; engineering took it to production.
In production, human-reviewed
Used by teams at
✦FinOps Agent
parsing request…
Cost Analyzer
Anomaly Detection
Optimizer
Report Builder
The story
Founder to AI-native PM, five chapters
Each move was deliberate — a harder data problem than the one before it.
2020 · The founder
I started by starting a company
Straight out of IIT Kharagpur — dual CS degree in hand — I founded BlockDeliver, a decentralized CDN marketplace. Blockchain incentives, spare bandwidth, partnerships with Kingsoft Cloud and ArvanCloud, a 15-person team recruited across five countries. It taught me the whole stack of building: strategy, architecture, hiring, and shipping an MVP with no safety net. It also taught me what I wanted next — to learn product craft properly, inside companies solving hard data problems.
2021 · The apprenticeship
Learning ML products at enterprise scale
At Soroco I worked on Scout's work graph — the ML that reconstructs how work actually happens from billions of raw activity signals, used by 200+ organizations including Fortune 500s. My job was making that reconstruction more accurate. It was the best possible apprenticeship in a truth I've relied on since: the model is never the product; the product is what the model makes visible.
2022 · The systems years
Two years inside how companies plan
At Drivetrain I owned planning and forecasting — moving finance teams from static annual plans to rolling forecasts, scenario modeling and what-if analysis, plus collaborative headcount planning tied to live P&L. Finance software is unforgiving: numbers must reconcile, and every user is an expert. It sharpened my respect for correctness and for deterministic systems — knowing when NOT to use ML is an AI product skill.
2024 · The AI-native turn
Building the memory of a company's customers
At Enterpret I worked end to end on the customer knowledge graph and adaptive taxonomy — the system that turns millions of scattered feedback records into structured, queryable intelligence, adopted by 50+ enterprise customers including teams at Notion, Canva, Linear and Perplexity. This is where my work went fully AI-native: continuous-learning classification, graph data models, and launching AI-powered win/loss insights I discovered and shaped from concept.
2025 · Now
FinOps for the AI era
At Amnic I'm a Senior PM across the portfolio: AI Token Management (LLM spend intelligence adopted by 10+ enterprises), Cloud Inventory (the multi-cloud system of record that helped win the company's largest six-figure deal), Virtual Tags (~75% customer adoption in a month), and AI-native analytics agents — which I prototyped myself in Claude Code and Figma Make before engineering took them to production. The through-line of my whole career, compounding: hard data problems, made legible, shipped honestly.