Shubh Sankalp DasSSD
Bengaluru--:--ISTResume ↗Let's talk

Shubh
Sankalp Das

BengaluruLinkedIn X GitHub

Currently
Product Manager at Dezerv
lending, voice agents
Building
Voxikin
after hours, now
Shipped
ToolMonkey, Filtr
case studies below
See the builds Let's talk

Drag anything on the desk. Click to open it.Swipe the desk. Tap anything to open it.

Shubh Sankalp DasShubh Sankalp DasProduct Manager, Dezerv
Lending + Voice AgentsToolMonkeyChaos testing for agent tool calls.RUN C14 MODESnoneokwrong_answermissedmalformed_jsonmissedsilent_failurecaughtHEALTH72.5/100FiltrTurns Slack, Jira and call notes into a ranked list of user problems.BlogP99 is a UX metric.BITS Pilani, GoaElectronics + FinanceVoxikinA voice AI elderly healthcare assistant.

Built and shipped, end to end.

2 shipped · 1 being builtCase study and live link for each#projects

ToolMonkeyChaos Monkey for LLM Tool-Calling Agents

First test run

100%tasks completed
33%failures caught
25%went unnoticed

Breaks your agent's tools on purpose, then shows exactly how it fails, before your users do.

AgentEval · LLM · FastAPI · Python

Read the case study →toolmonkey.vercel.app ↗
ToolMonkeyReliability report
Scenario C1
What is 847 × 23?
wrong_answer
tool19,841agentanswered 19,841, no flag
MISSED
malformed_json
tool{"res": 1948agentanswered, no flag
MISSED
silent_failure
tool(empty)agentflagged missing data
CAUGHT
Health score72.5 AMBER
72.5AMBER

FiltrRAG-Based Insight Engine for Product Managers

What you get

~45 sto first insight
5ranked issues
3sources

Reads your call transcripts, Slack and Jira tickets, then ranks the biggest issues by impact so you know what to fix first.

RAG · LLM · FastAPI · Pinecone

Read the case study →filtr-omega.vercel.app ↗
Callcheckout keeps failing on my phone
Slackcannot pay on iOS again
Jira PAY-412Payment sheet crashes on Safari
Callthe export takes forever
Slackcsv export timed out again
Slackcan we change the report colors
1Mobile Checkout Failures
7 mentionsHigh impact
Fix first
3 sources, 1 answer

VoxikinVoice AI elderly healthcare assistant

Caring for your loved ones, no matter where life takes you.

It calls your parent on schedule, confirms every dose in their own language, and tells you only what actually needs you. They install nothing. Their entire interface is answering a phone call.

Not public yetAsk me about it →

Product work, measured.

4 companies · 2024 to nowFull-time and internships#experience

Dezerv

Bengaluru · 2 roles · '25 to now
Full-time · Bengaluru

Product Manager

Lending productsLoan against mutual funds and loan against securities.Voice AI agentsFor PMS client calls.

Jul 2025 to Dec 2025 · Internship

Product Intern

38→51%M1 retention

by conducting user research on weekly portfolio tracking needs and shipping the Portfolio Snapshot feature.

  • 94→2%status misclassification
  • 18%more conversions, same RM capacity
  • +40%session time, controlled cohort
All 4 results
  • M1 retention lifted from 38% to 51% by conducting user research on weekly portfolio tracking needs and shipping the Portfolio Snapshot feature.
  • Status misclassification reduced from 94% to 2% by engineering a client call status identifier that analysed transcripts using classification rules.
  • 18% more conversions with the same RM capacity by developing an affluent user model using salary, spend, and investment patterns.
  • Session time increased 40% in a controlled cohort by discovering the need for actionable insights and integrating Thurro AI into the stocks page.

91Ventures

Remote · 1 role · '24 to '25
Apr 2024 to Apr 2025 · Internship

Investment Analyst Intern

97.9%less effort on quarterly LP reporting

(48 hours to 60 minutes) by programming Excel automation for the reporting workflow.

  • 83%portfolio review time cut
  • 50%deal closure time cut
All 3 results
  • Portfolio review time reduced by 83% (30 to 5 minutes) by consolidating MIS reports into curated summaries for 200+ LPs.
  • Quarterly LP reporting effort cut by 97.9% (48 hours to 60 minutes) by programming Excel automation for the reporting workflow.
  • Deal closure time cut by 50% (60 days to 30 days) by establishing standardised deal documentation and communication workflows.

Multigraphics

Delhi · 1 role · '24
May 2024 to Jul 2024 · Internship

Business Analyst Intern

3xdealership inquiries in 30 days

(33 to 95) by identifying user drop-off due to lack of product context and redesigning the inquiry flow.

  • 1.8%conversion rate, from under 1%
  • 92→27%bounce rate in 45 days
All 3 results
  • Dealership inquiries increased 3x in 30 days (33 to 95) by identifying user drop-off due to lack of product context and redesigning the inquiry flow.
  • Conversion rate increased from under 1% to 1.8% in 60 days by analysing data identifying delivery drivers as the highest-intent cohort and restructuring outreach.
  • Bounce rate reduced from 92% to 27% within 45 days by restructuring homepage CTAs with product categories and test drive options.

Product Space

Remote · 1 role · '24
Mar 2024 to Apr 2024 · Internship

Product Intern

+15%signups within 2 weeks

by restructuring the website with testimonials and sample decks to address the trust barrier, validated via A/B testing.

  • 22%higher conversion than other sources
  • 25 to 30%lower cost per signup
All 3 results
  • +15% signups within 2 weeks by restructuring the website with testimonials and sample decks to address the trust barrier, validated via A/B testing.
  • 22% higher conversion rate than other traffic sources achieved by releasing a monthly Product Newsletter to drive cohort awareness.
  • Lead attribution time cut by 1 week, cost per signup reduced 25 to 30% by implementing a UTM-linked conversion dashboard tracking 9 sources.

Thinking in public.

1 published · 1 in draftBlogs on the product decisions inside engineering#writing

LLM Infrastructure · Oct 2026

P99 is a UX Metric. Uber Just Hid It in an SLO.

This is why Uber did not optimize for average latency. They optimized for the tail. Because the tail is where trust breaks.
6 min read6 sections
  1. The Setup
  2. What P99 Actually Measures
  3. The System Got Smarter. Then It Got Unpredictable.
  4. The Decisions That Were Actually Product Decisions
  5. What PMs Miss When They Read Infrastructure Posts
  6. The Takeaway
Read the blog →Each bar is a section. Pick one to jump in.

The person behind the work.

#about
The most interesting product problems are people problems.

I studied Electronics and Instrumentation at BITS Pilani Goa, with a Finance minor. I drifted into product because my questions were never only about technical output. They were also about people. Now I'm a Product Manager at Dezerv.

Outside work I train, read history and mythology, and travel when I can.

Loading contributions…github.com/shubhdas0208GitHub ↗

Each square is one dayLess More
Obsessing overF1: Race Strategy
0:00.00

How I work

  1. Ask users firstPortfolio Snapshot →
  2. Break it before users doToolMonkey →
  3. Watch the tail, not the averageP99 is a UX Metric →

Me between the Pages 74 books · pull one off the shelf

    Reading now
    Product & Tech
    Startups & Sales
    History & Politics
    Lives & Mind
    Sci-fi
    Mythology
    Myth Thrillers
    Thrillers
    Percy Jackson
    Heroes of Olympus
    Kane Chronicles
    Trials of Apollo
    Magnus Chase

Need a PM who ships?Let's talk.

Form, email, LinkedIn or XBengaluru · IST#contact
dasshriyans2802@gmail.com

It's --:-- in Bengaluru right now.

Shubh Sankalp DasProduct Manager at Dezerv · Bengalurulinkedin.com/in/shubhsankalpdasConnect Shubh Das@shubh_das02x.com/shubh_das02GitHubshubhdas0208Loading contributions…ResumeOpens in Google DriveRead my resume