AI
Leading AI in the SDLC
How I lead teams from AI-assisted to AI-orchestrated delivery - with governance, quality, and measured outcomes.
The through-line
I lead a ~30-engineer multi-location engineering across enterprise platforms and still work hands-on across architecture and Agentic AI workflows. The goal is consistent: make Human-AI collaboration disciplined enough to measure, not demo theater. For a senior engineering leader, the real question is not "do you use AI" - it is how you govern it, keep quality high, and prove the outcomes.
How I lead
Engineering leadership here means building the org, not only the model. I grew a multi-location team from 15 to 22 to ~30 engineers across Noida and Mysore, with 9 direct reports, managed through team leads. Scope covers hiring and mentoring, delivery governance, stakeholder management, roadmap ownership for a multi-product portfolio, and ownership of hiring plans, attrition, promotions, and engineering budget / cost decisions. Hands-on work stays in the mix: Pearson CRX microservices, shipped AI product features, and a privacy-first self-hosted LLM platform - while leading 9 directs. The same bar applies to AI adoption: I still ship architecture and quality gates while the team scales, and keep product contribution and research in the mix.
Two separate tracks
Excel AISprints - proprietary Excelsoft Excel AI methodology. Human-AI collaboration built around technical PRDs, guardrails, and micro-sprints, practiced on the ExcelAI Platform. This is where my structured AI-assisted delivery work began, with measured pilot outcomes. Employer practice - not the public kit.
Agentic SDLC - a public orchestration approach and MIT kit for AI-first delivery: intent briefs, context gates, optional multi-agent swarms for multi-unit increments, eval-driven quality gates, and the Delegate / Review / Own loop. Aligned with AWS AI-DLC in operating model; not an AWS implementation and not Excelsoft product IP.
Excel AISprints / ExcelAI are Excelsoft. Agentic SDLC is the downloadable, AWS-DLC-based template anyone can explore.
Self-hosted LLM governance
Privacy-first self-hosted LLM tooling (Mistral / Codestral family) for coding, test generation, and documentation, used by ~150 internal engineers at Excelsoft. Models change as stronger open-source options appear. The priority is governance, security, and real adoption.
Agentic product quality and ML
Production machine learning in EdTech: student retention and course-risk models at ~91% precision on millions of learning events, alongside agentic remediation patterns kept inside strict guardrails. Deeper pedagogy, stack breadth, and AI-in-learning stance live on Learning Engineering.
Measured outcomes
On Excel AISprints pilot feature releases at Excelsoft: ~50% defect reduction and ~30% engineer time saved where AI-assisted flows applied. These are pilot results, labeled as pilots. See Results & impact for the verified numbers in context.
How teams adopt the public kit
A phased, lower-risk rollout: context stabilization, pilot micro-sprints, automated eval integration, then standardization. Public kit lives under Agentic SDLC - also listed under Build.
Who it is for
Engineering leaders adopting AI in the SDLC who still care about quality, ownership, and measurable delivery - especially in regulated or high-scale product organizations.