Technical ownership
Translate model, infrastructure and partner requirements into a working decisioning platform.
TECHNICAL FOUNDER · AI PRODUCT & PLATFORM BUILDER · GOOGLE FOR STARTUPS ALUM
I build AI products
from zero to scale.
I co-founded KarmaLife and led product, engineering and applied AI/ML from inception to a lender-funded platform serving 2M+ users. I combine hands-on technical judgment with product strategy, team building and enterprise execution.
Before KarmaLife, I spent 12 years at Samsung R&D working across computer vision, biometrics, multimedia and mobile systems. Today, I focus on building AI-native products and platforms: finding the right problem, shipping what works, and scaling with quality, reliability and sound economics.
20+ years in engineering and AI · Co-Founder & CPTO · Google for Startups Accelerator alumnus · IEEE Senior Member
2M+users on a platform I personally architected
96%underwriting automated, ~12s → ~1s decisions
₹300–350 Crannual disbursements (~$32–36M)
30core team org built from zero
Founder, engineering & ecosystem evidenceGoogle for StartupsVenture-backed founderIIMA Ventures mentorTechnical educator
0→1 product ownership
Co-Founder & CPTO
Built and scaled a venture-backed startup from zero to 2M+ users.
Production AI & data
Cloud-native + applied AI
Kafka · Kubernetes · Golang · ML evaluation · GPU pipelines · LLM workflows.
Computer vision roots
12 years at Samsung R&D
Computer vision · biometrics · multimedia · international patents.
Enterprise product leadership
Product · Engineering · ML · GTM
Built teams and aligned enterprise customers, lenders and partners around measurable outcomes.
01 / Built at scale
Architecture is only useful
when the business works.
At KarmaLife, I owned product and technology end to end. The challenge was to make credit decisions fast, explainable and trusted enough for regulated lenders to fund them.
Turning a two-sided deadlock into a platform.
India's gig and blue-collar workers earn steadily but have no credit file, so bureau-based lenders reject them by default. KarmaLife built the decisioning layer to underwrite them anyway and ran straight into a deadlock on both sides.
regulated lenders
won't fund without proof the model holds
deadlock
gig marketplaces
won't integrate without a lender ready to disburse
broken by making the lender's risk team the first customer, sharing model performance and governance openly, until trust closed the deal
My role: Co-Founder and Chief Product and Technology Officer. Architecture, product decisions, hiring and leadership across engineering, data science, product and risk.
Read the HyperVerge customer case study ↗Production architecture / KarmaLife
From signals to a credit decision.
Event sourcesapp · bureau · device
Kafkastreaming pipelines
Kubernetes + Golangorchestrated microservices
MongoDBstate & features
ML decisioningXGBoost + SHAP
Credit decision~1s, explainable
High-level architecture of the platform I personally led, at 2M+ users. The lesson was not the specific stack. It was knowing what had to be event-driven, what belonged in synchronous decisioning, where state should live, and which complexity actually mattered at each stage of growth. Production outcomes are separate from the research projects below.
Executive trust
Explain performance and governance in the language of lender and risk stakeholders.
Cross-functional execution
Bring product, engineering, data science and risk together around partner constraints.
Influence without authority
Aligning teams I did not control.
As a technical founder, I made architecture and product decisions across very different constraints, from marketplaces and lenders to risk, investors, and my own product, engineering and data teams.
My role was to turn those competing constraints into clear architecture, product and operating decisions that everyone could support.
Marketplaceintegration speed
Lendercontrolled credit risk
Engineeringreliability and scale
Productconversion and usability
Riskexplainability and governance
02 / Product building
Hands-on judgment.
Enterprise-scale outcomes.
I have owned the journey from early product definition and technical architecture to ML decisioning, customer integration and cross-functional execution. I build the systems and the teams that turn difficult problems into repeatable products.
Zero-to-one execution
Turn customer constraints into a clear product bet, build the initial architecture, ship, learn and iterate.
Production AI and data
Move beyond model demos with reliable pipelines, meaningful evaluation, explainability, monitoring and human judgment where needed.
Enterprise product leadership
Align product, engineering, ML, risk and enterprise partners around outcomes, adoption and operational trust.
03 / Founder decisions
The calls that turn
ideas into platforms.
From customer discovery and architecture to model governance, the hard decisions shape what a product can become. Three examples from building KarmaLife.
Who is the first customer?
- Constraint
- Lenders wouldn't fund without proof; marketplaces wouldn't integrate without a lender.
- The call
- Treat the lender's risk team as the first customer. Open up model performance and governance before asking for capital.
- Trade-off
- A slower first deal and deeper scrutiny of the model, in exchange for trust that compounds.
- Result
- A lender-funded platform disbursing ₹300–350 Cr a year.
What actually has to be real-time?
- Constraint
- Slow decisions cost conversion inside partner apps, and the platform had to keep scaling.
- The call
- Move signal ingestion into event-driven pipelines and keep the synchronous path to the decision itself.
- Trade-off
- More moving parts in the data layer, in exchange for a fast, predictable decision path.
- Result
- ~12s → ~1s decisions, 96% of underwriting automated.
Where does generative AI belong?
- Constraint
- Regulated decisions must be explainable and reproducible; LLM output is neither by default.
- The call
- Keep the decision deterministic and explainable. Use LLMs for interpretation, audit summaries and operations around it.
- Trade-off
- Less "AI magic" in the pitch, more confidence from the people who sign off.
- Result
- AI that risk and compliance teams can audit, not just demo.
04 / AI systems: from models to production
From models
to production systems.
I connect product questions to hands-on engineering: prototyping approaches, evaluating models and data quality, and designing systems that work beyond a demo. Clear evaluation, observability and operational reliability matter as much as model accuracy.
US credit-risk
model evaluation
Logistic regression vs. XGBoost · SHAP explainability
Compare logistic regression and XGBoost on SBA 7(a), HMDA, Freddie Mac and Fannie Mae data. Examine TreeSHAP explanations, a later-period holdout and calibration gaps rather than treating ranking performance as decision readiness.
Python · scikit-learn · XGBoost · TreeSHAP
Explore the repository ↗GPU processing +
LLM audit summaries
XGBoost · 0.869 AUC
An underwriting workflow combining GPU dataframe processing, predictive models and SHAP adverse-action explanations with an LLM-generated audit-summary layer. The summary layer supports interpretation; it is distinct from the predictive model.
RAPIDS cuDF · XGBoost · NVIDIA NIM · Llama
Explore the repository ↗Alternative-data
credit infrastructure
0.826–0.893 AUC · 4 markets
One alternative-data underwriting engine for India, Indonesia, the Philippines and the US, generating regulator-specific loss models for IFRS 9, Ind AS 109 and US CECL from config, not code.
Alternative data · ML · Loss modeling
Explore the repository ↗I keep exploring new ideas through research and prototypes, testing models, architectures and AI workflows beyond production systems.
05 / Founder ecosystem
Make the technical
conversation useful.
I have spent the last three-plus years mentoring founders, teaching, speaking and contributing to startup ecosystems alongside building KarmaLife.
IIMA Ventures + Startupbootcamp
Mentor across founder-focused startup ecosystems, including IIMA Ventures office hours and Startupbootcamp's Mentor Meet Startups program, working with DeepTech & Robotics founders through 1-on-1 sessions.
Cloud architecture and GenAI economics
Roundtable participation on cloud-driven workflow efficiency, generative AI return on investment, and harnessing generative AI for enterprise workflows.
AI, data and enterprise adoption
Ecosystem discussions with Oracle, Confluent, Snowflake, Skyflow and Wiz. Judging for startup and enterprise awards.
Selected technical writing
AI systems need execution discipline.
Building an underwriting agent: reasoning and enforcement ↗AI agents and execution control ↗Reliable underwriting systems at scale ↗
Selected recognition
Recognised for building real.
NASSCOM India Fintech Day 2023 award for KarmaLife · Financial Express "FE Visionary Leader" · World Bank CGAP pilot partner and featured case study · international patents from Samsung R&D.
Also in the room for
Inc42 AI Summit (speaker) 2026
AI Summit: How enterprises are re-architecting data for the agentic AI era (speaker) 2026
Outlook Hub4Business profile 2026
Confluent AI-powered data insights and real-time business operations (speaker) 2024
Oracle AI-powered business operations and data management (speaker) 2024
Snowflake AI and data in fintech (speaker) 2024
Salesforce personalization at scale, AI/ML and cloud analytics (speaker) 2023
Financial Express FinTech Summit (speaker) 2023
World Bank CGAP (featured case study), transactional-data lending 2024
World Bank CGAP (selected pilot partner), platform-worker credit scoring 2022
BFA Global Catalyst Fund (7th cohort), inclusive fintech 2020
06 / Founder perspective
From zero to production.
From product to business.
I turn complex engineering and AI opportunities into products people and enterprises depend on: discover the constraint, build to prove value, measure outcomes and scale what works.
Architecture & scale
Data boundaries, synchronous vs. asynchronous execution, observability, scaling and build-vs-buy without imposing enterprise complexity before it is needed.
AI economics
Where generative AI creates real workflow leverage, and how model choice, routing and caching shape cost, latency and margin as usage grows.
Agents & execution control
Where an agent should reason and where it must be constrained: tool boundaries, guardrails, human review and failure handling before it touches customers.
Evaluation before scale
Turning "the demo works" into measurable quality: evaluation sets, holdouts, calibration and monitoring that tell you when a model is decision-ready.
Enterprise & regulated customers
Translating security, data, explainability and governance requirements into product and architecture choices when a bank, insurer or enterprise becomes the customer.
Technical diligence
Preparing the architecture, data and AI story that investors, acquirers and enterprise buyers will probe, and fixing what won't hold up.
My perspective spans the full loop: founder and product owner, hands-on AI and engineering leader, and enterprise customer partner. The goal is always reliable technology tied to measurable business value.
07 / Engineering foundations
Depth before
the founder title.
My career spans embedded systems, mobile platforms, computer vision, biometrics, applied ML and enterprise fintech. That breadth helps me connect implementation details with product strategy, customer needs and business outcomes.
Education
B.E., Electronics and Communication
Karnatak University
Senior Management Programme
Indian Institute of Management Ahmedabad
B.E., Electronics and Communication
Karnatak University
Senior Management Programme
Indian Institute of Management Ahmedabad
2019–present
KarmaLife
Co-Founder · Chief Product and Technology Officer
Venture-backed fintech. Product and technology ownership from initial development to a platform serving 2M+ users.
2018–2019
Smartron
Director · AI Labs
Established and led applied AI and deep-tech initiatives within the tronX platform.
2006–2018
Samsung R&D
Senior Technical Manager
Worked across mobile, computer vision, biometrics, multimedia and security; built engineering teams and contributed international patents.
2001–2006
Early engineering roles
Celstream · Sasken · Pinexe · Suvistas
Software Engineer to Team Lead in embedded systems and multimedia codecs.
08 / How I build
Own the problem.
Build what matters.
My operating approach connects customer discovery, first-principles technical choices, fast execution, reliable delivery and continuous feedback across product, engineering and enterprise go-to-market.
Discover the actual constraint
Talk directly to customers and stakeholders; prioritize the problem worth solving before committing to an architecture.
Make technical trade-offs explicit
Choose the right data, model, architecture and build-vs-buy boundary for the stage, without complexity for its own sake.
Ship and measure
Translate ambiguous opportunities into working products, evaluations and business metrics rather than just specs or demos.
Build the team and improve the system
Create ownership across product, engineering, ML and commercial teams; feed production evidence into the next iteration.
My build-to-scale loop
01
Discover
Talk to customers early and define the real job, constraint and measurable outcome.
02
Design
Set crisp product bets and technical boundaries. Make reliability and costs explicit.
03
Build
Prototype, write code where it adds leverage and lead a focused team to ship.
04
Scale
Make quality, observability, security and operations repeatable as adoption grows.
05
Iterate
Use model evaluation, customer usage and commercial outcomes to guide the next release.
From the founder's side of the table
What founders need from the platforms they build on.
As a founder, I was the customer of cloud, AI and data platforms. These are the gaps I felt most, and the ones I hear repeatedly from founders I mentor.
- Predictable AI economics past the free creditsFounders need to see what inference will cost at 10× usage before they commit an architecture to it.
- Capacity at the moment of launchQuota and accelerator access matter most in the week a product takes off, not in the procurement cycle after.
- Reference patterns for evaluation and governanceNot more demos, but proven ways to test, monitor and explain AI systems that enterprise and regulated buyers will accept.
- One technical person who knows their stackA single point of contact who understands the architecture and can route problems to the right team quickly.
I've experienced the value of the founder ecosystem.
Google selected KarmaLife for its AI/ML Accelerator, one of 20 startups chosen from more than 1,050 applications.
Read Google's cohort announcement ↗