Manvendra Singh Tanwar · Product dossier · Doc № 2026-01
0%
Manvendra Singh Tanwar · Product at Visa2Fly In production

Evidence, not adjectives.

Product at Visa2Fly, the $10M ARR B2B visa platform backed by Flipkart Ventures, M Venture Partners, and FinSight Ventures. I ship four surfaces: the AI visa engine, the multi-tenant partner platform, the 65-country consumer journey, and the org-wide experimentation system. Every claim on this page has a number attached.

See the evidence ↓ Resume LinkedIn ↗
Open to work
Manvendra Singh Tanwar, photographed straight on against a plain wall
Exhibit A

Measured outcomes.

Five numbers from four years of shipped work, FY 2022 to 26. Each one is auditable; ask me about any of them.

A.1 Throughput on the AI visa product, same ops team, same infrastructure ×
A.2 Turnaround compression on the AI eVisa journey, ~15 hours to ~1 hour %
A.3 Cost to file each eVisa, ~$1.00 to ~$0.25: the unit economics that fund the 65-country roadmap %
A.4 Active B2B agents, from a base of 16. B2B grew from ~25% to ~75% of business +
A.5 Seed round closed, data room owned end to end $M
§01 · Changelog

Four years, three lenses, one problem.

Joined Visa2Fly as employee #11. Ops first, then analytics, now product.

I prefer the messy middle of 0-to-1: writing the PRD, sitting next to the AI engineer, watching the funnel the day after launch. I trust SQL more than vibes.

BaseGurugram, India · open to relocate
EducationBBA, GGSIPU Delhi · 8.8/10 GPA
StanceOperator first
v3.0 / current

Associate Product Manager

Nov 2025 to present

Run product across four surfaces: AI visa engine, multi-tenant partner platform, 65-country consumer journey, experimentation system. Report to the co-founder (CTPO).

v2.0

Business Analyst

Nov 2022 to Oct 2025

Built the org's experimentation system, the unified Sales and Ops product, CAC/LTV analytics, and the data room that closed the $2M seed.

v1.0

Operations Manager

Jun 2022 to Oct 2022

Founding-team operator. Stood up vendor onboarding and SLA frameworks; led city expansion across Chennai, Bangalore, Hyderabad, Mumbai. 35% CAGR post-launch.

§02 · Work

Work, itemised.

Nov 2025 to present
Gurugram
In production

Associate Product Manager

Visa2Fly · Multi-tenant B2B SaaS platform · $10M ARR

  • Shipped the AI eVisa product end to end, ideation to launch: 4× throughput on the same ops team, cost to file each eVisa cut from ~$1.00 to ~$0.25, turnaround from ~15 hours to ~1 hour, fully automated across the 10+ countries that drive ~85% of volume. Unit economics now support the full 65-country roadmap.
  • Scaled the B2B partner platform from 16 to 300+ active agents (B2B grew from ~25% to ~75% of business), lifting monthly repeat-active partners from ~5% to ~50%: self-serve dashboard with bulk upload (100+ travelers, sub-15s auto-mapping), SLA timers plus API, activation flow, customer success motion.
  • Shipped the Sticker Visa product across 65+ countries: authored the PRD, coordinated CTO, AI engineer, and design from spec to launch. 95% TAT compression, headcount neutral.
  • Own the OCR training-data strategy and confidence-based escalation thresholds. Human-in-the-loop review held under 2% of cases.
Nov 2022 to Oct 2025
Gurugram

Business Analyst

Visa2Fly · Analytics, experimentation, growth instrumentation

  • Built Visa2Fly's experimentation system: score-ranked routing vs. round-robin, 2,000 leads per arm. 22% conversion lift, 15% lower CAC at statistical significance. Adopted org-wide.
  • Shipped a unified Sales and Ops product for the 50-person team, replacing three disconnected systems. 5 SLAs instrumented with breach alerts. 27% faster customer connect.
  • Built CAC/LTV and funnel analytics in SQL, Python, Tableau. Sunset underperforming channels: 50% marketing burn cut, 18% ROI lift.
  • Ran the $2M seed data room end to end: financial model, cap table, ESOP pool, investor packs. Closed with M Venture Partners, Flipkart Ventures, FinSight Ventures.
Jun 2022 to Oct 2022
Gurugram

Operations Manager

Visa2Fly · 11-person founding team

  • Built operations infrastructure from scratch: vendor onboarding, SLA framework, process documentation, field-agent coordination.
  • Led city expansion into Chennai, Bangalore, Hyderabad, and Mumbai as EIR. 35% CAGR post-launch.
§03 · Featured product

The AI visa engine.

Runs end to end on AI: 4× throughput on the same team, turnaround ~15h to ~1h, cost to file each eVisa ~$1.00 to ~$0.25. Five stages, each with a measured gate. Fully automated in 10+ countries, the ones carrying ~85% of volume. Work is live on the rest.

01OCR extraction
02VLM validation
03Bot filing
04Tokenised payments
05Post-approval
01 / Intake

OCR extraction

Document intake runs through a tuned OCR layer that pulls passport, photo, and form fields from scans and phone photos, even crumpled ones. Fields normalise into a schema the rest of the product trusts.

98%
Field accuracy
<3s
Per document
OCRSchema normalisationImage preprocessing
02 / Quality gate

VLM validation

A vision-language model verifies every OCR pull: matches passport photo to form, flags mismatched names, spots blurry pages, reads handwritten edits. Anything below confidence routes to a human before we touch a government portal.

2%
Escalation rate
10+
Countries fully automated · 85% of volume
VLMConfidence thresholdsHuman-in-loop
03 / Execution

Bot filing

Automated bots file the validated application into each country's portal, handling session state, captchas, and varying form shapes. What took agents 20 minutes of copy-paste now finishes in under 3.

20m → 3m
Filing time
4×
Throughput, same team
Bot automationSession managementPortal integrations
04 / Revenue

Tokenised payments

Payment orchestration triggers at the right handoff: government fees, service fees, auto-refund on rejection, with reconciliation back to finance. No stuck applications waiting on a payment ticket.

0
Manual reconciliations
100%
Auto-refund on reject
TokenisationReconciliationFinance integration
05 / Delivery

Post-approval

Once a visa is granted, the product downloads the approval, packages it with travel docs, notifies the customer, and closes the loop in CRM. SLAs go red the moment something slips.

5
SLAs tracked
Real-time
Alerting
SLA designNotificationsCRM close-out
§04 · Case files

Seven shipped case files.

Click any row for the one-paragraph version. Four link to a wireframe or an architecture diagram. Three are internal systems, so those stay a conversation.

F.01 B2B Partner Growth Engine 16 → 300+ agents · ~75% of business · repeat-active ~5% → ~50% Open ↗ Owned the B2B growth motion end to end. Four levers: self-serve dashboard with bulk upload (100+ travelers, sub-15s auto-mapping) and live tracking, SLA timers plus API, activation flow, customer success motion. B2B grew from ~25% to ~75% of business. F.02 Customer Onboarding Flow 8 → 5 steps · doc validation ~48h → ~2h · 25+ interviews Open ↗ Redesigned the consumer visa journey from 8 manual steps to 5 automated ones and cut document validation from ~48 hours to ~2. Authored the PRD from 25+ customer interviews and journey mapping, in partnership with design. F.03 Partner Dashboard Multi-tenant · RBAC · 100% SLA met · bulk 100+ travelers Open ↗ The operator surface agencies run on: real-time queue with SLA timers, AI-flagged items first, bulk CSV ingestion, live webhook stream, API usage tracking. F.04 AI Workflow Engine 98% accuracy · filing 20m → 3m · 2% human-in-loop Open ↗ End-to-end automation: doc matching, cleanup, cross-validation, instant cover letters and itineraries, auto-filing, live status, multi-channel delivery. Powers the platform's unit economics.
F.05 Experimentation System +22% conversion · −15% CAC · at significance Detail ↓ Score-ranked routing vs. round-robin, 2,000 leads per arm, significance in 4 weeks. Now the org default for funnel decisions across product, sales, and growth. The test design, guardrail metrics, and the two results that went the wrong way are a walkthrough, not a wireframe. Ask me for it.
F.06 $2M Seed Data Room $2M closed · 3 VC firms · model, cap table, ESOP Detail ↓ Co-owned the investor data room end to end: financial model, cap table, ESOP schedules, reporting packs, Q&A. Closed with M Venture Partners, Flipkart Ventures, FinSight Ventures. Nothing from this room is public, and it should not be. Happy to talk through how it was structured.
F.07 Sales & Ops CRM 3 tools → 1 · −50% marketing burn · +18% ROI Detail ↓ Unified internal product for the 50-person team. Data model, 5 SLAs with breach alerts, CAC/LTV instrumentation in SQL, Python, and Tableau. Internal tooling, so the screens stay internal. The data model and the SLA design are the interesting part and I will walk through both.
§05 · Writing

How I write a spec.

Two documents on a system I built and still own, written independently of my work product. Volumes, unit costs, and vendor terms are rounded or left out. Six pages and four. Open either one.

W.01 PRD · Visa Document Engine v2 6 pages · per-field confidence · staged 12-week ask Open ↗ The spec for the next version of the engine I shipped. Names the three things that stop it scaling, sizes the opportunity with the assumption printed next to every number and one marked "to confirm", and replaces a single confidence threshold with one per field weighted by what that field costs when it is wrong. Includes what I am deliberately not building, the failure I actually worry about, and the four questions I still cannot answer. W.02 Eval Plan · Document Extraction 4 pages · 99.5% on dangerous fields · release gate Open ↗ Why "96% accurate" is a useless number, and what replaces it. A 1,500-document test set weighted towards cases that already broke us, accuracy reported per field and per country instead of as one average, a calibration check asking whether the confidence scores mean anything, and the six gates a release has to clear before it ships.
§06 · Capabilities

What I bring.

Product
PRDs & specsRoadmappingExperimentation designA/B testingUnit economicsUser researchFigmaRICEGo-to-marketOKRs
AI surface
OpenAI APIsOCR & document AIVision-language modelsHuman-in-the-loop designConfidence thresholdsPrompt evaluation
Data
SQL, advancedPythonTableauPower BIMixpanelGoogle AnalyticsData modelingKPI dashboards
Execution
Stakeholder managementCross-functional leadershipSLA frameworksInvestor relationsJiraNotionClickUp

Hiring for product? Open the file.

Hiring an APM with operator range? I'd love a 20-minute chat. Role, collaboration, or coffee about AI products: all welcome.

Emailmanvendrasinghtanwar.work@gmail.com Phone+91 93553 15577 LinkedInmanvendrasinghtanwar ↗ ResumePDF, one page ↓
LocationGurugram, IN · open to relocate