AI systems that can be evaluated
Define evidence boundaries, schemas, quality gates, approval points, and failure behaviour before an AI workflow reaches users.
See evidence: Scoped agent access PoCEngage: AI Production Readiness
Somesh Jha
Build the right system. Get it into production.
I help technology leaders make consequential architecture decisions, modernize complex platforms, and move AI systems from prototypes into controlled production.
01 / About
Architecture is useful only when teams can deliver it.
I’m Somesh Jha. I lead senior architects and advise enterprise clients across banking, insurance, and telecom on the technology decisions that determine whether a platform actually reaches production.
Over the past 3.5 years I’ve grown the architecture and engineering practice I lead from roughly 3 to 54 people, operating through senior architects and distributed teams across banking, insurance, and telecom. That scale matters: knowing how to architect a distributed platform is one skill; knowing how 40+ engineers actually deliver it is another, and enterprise clients need both.
Across 13+ years of public work and generalized client engagements, I’ve led complex technology initiatives from discovery through delivery: modernizing monolithic applications, integrating mainframe systems with Kafka, defining cloud transformation roadmaps, and developing computer-vision and AI platforms.
DecideMake ownership, trade-offs, and constraints visible.
DeliverTurn the target state into sequenced work teams can ship.
OperateDesign for release, support, compliance, and change, beyond what it takes to get approved.
02 / Services
Where I can help
Hands-on advisory for organizations making consequential platform, modernization, and AI decisions.
Architecture tied to delivery
Turn business constraints into target-state designs, decision records, delivery increments, and named technical ownership.
Read: Building an architecture decision systemEngage: Architecture Decision Sprint
Incremental modernization
Find viable seams in legacy systems, sequence migrations, and keep customer and operational flows working throughout the transition.
See evidence: Tier-1 banking & insurance modernizationRead: Modernizing a legacy core
Critical design reviews
Test assumptions across data, integration, security, operations, and compliance; record the decision and the risks that remain.
Technical counsel for leaders
Give executives and clients an independent view of platform choices, delivery exposure, team capability, and where investment will matter.
03 / Selected work
Systems I have built and led
Public work and generalized client engagements. The details below are specific; client-sensitive information is not.
Archavow architecture workbench
A local-first application that turns incomplete requirements into HLDs, C4 diagrams, ADRs, risks, and backlog items while keeping every AI-assisted step reviewable.
- 270+ automated tests
- 3 CI pipelines
- Next.js, FastAPI, PostgreSQL/pgvector
Real-time data marketplace
Designed and delivered the first release of an event-driven integration platform, then led architecture and delivery for the second release across client teams and vendors.
- Kafka integration
- Cross-vendor delivery
- Production rollout
Group insurance modernization
Assessed the legacy estate, designed a cloud-native target state, and facilitated the engineering and executive decisions needed to approve a multi-year roadmap.
- Legacy assessment
- Azure target state
- Executive decision forums
Tier-1 banking platform modernization
Architecture and engineering leadership for high-volume financial platforms modernizing product, pricing, and account processing workloads, delivered through senior architects and distributed teams.
- Architecture governance
- Distributed team leadership
- Production-scale delivery
Also: named inventor on US Patent 11,270,105 for extracting and analyzing information from engineering drawings.
04 / Let’s talk
Have a difficult technology decision?
Most of my work starts with a question rather than a statement of work.
ArchitectureAre we making the right technical decision?
AI & AgentsCan we safely move this into production?
ModernizationHow do we change this system without creating another multi-year transformation?
EngineeringWhy isn’t this system scaling or delivering reliably?
If you’re working through something similar, let’s talk. No deck required, no sales pitch.
30-minute introductory conversation
Book 30 minutes. Bring the problem; no deck required.
Schedule a conversation
Prefer email? me@someshjha.com. I read and respond to these personally.