Opinderjit S. Bhella

Platform Leadership

Built an AI/ML platform organization from zero to 100+ engineers

As Head of MLOps & Software Engineering at HP, I built the enterprise AI platform organization from a founding team to 100+ engineers, managers, and contractors. I owned the $15M+ annual operating budget and presented AI platform strategy and investment roadmaps to SVP and C-level stakeholders. The business cases I authored won $13M+ in net-new multi-year program funding beyond that base. Keeping that team was as much the job as building it — structured career ladders, mentorship programs, and leadership development pipelines improved senior IC and engineering manager retention by 30%.

I built the MLOps platform itself from the ground up on AWS — owning the architecture, vendor negotiations, cost modeling, and executive alignment across business units. It supported 50+ production models across global regions with 99.95% uptime, under governance frameworks that kept 300+ workloads secure and compliant through automated policy enforcement and audit trails.

The recurring failure was not infrastructure. It was the handoff. Models reached QA with output fields added, dropped, or renamed, and every surprise meant another round of clarification, test updates, and retesting. I moved contract ownership to the data science teams: each model declared its input and output schema, shipped with its own tests, and was validated at runtime before a partner application ever called it. One model arrived with more than 300 inputs. Asked which ones it actually used, the team did not know. Over several months we cut it to fewer than 20. Model handoff and release dropped from roughly four weeks to four days, and partner QA could sign off in a single cycle.

On a separate program, I led the contractor team that moved our data engineering pipelines off a proprietary on-prem platform to AWS and Databricks, in four months against a six-month plan. We ran both environments in parallel for about a month and cut over only after the daily outputs consistently matched. There was no downtime at cutover, and finishing early returned two months of contracted engineering capacity to application work.

Product Leadership

Owned the roadmap for a consumer web platform built from scratch

Before building the platform organization, I spent roughly seven years at HP owning product backlogs and roadmaps for direct-to-consumer web platforms — three of them as product manager and senior product manager. HP Connected 3.0, which generated 21M+ monthly events, was built from scratch with a new co-located Scrum team; I owned the roadmap, backlog, and release scope, and served as Agile Product Owner through launch and the subsequent sunset of HP Connected 2.0.

That work meant arbitrating competing requirements across seven partner products into a single sequenced roadmap, running A/B tests in Optimizely against live Instant Ink subscription traffic, and owning the analytics instrumentation and user research that decided what shipped next. One of those tests settled Instant Ink's lead message: saving money on printing beat the environmental case for returning cartridges on completed subscriptions, not just clicks, and the gain held after rollout. I managed the OneTrust and GDPR privacy implementation and led the security reviews gating the platform's public launch across North America, EMEA, and Asia-Pacific.

It is the half of my background that makes the platform work land — I have sat on the side of the table where the roadmap gets cut, and I know what it costs a product team when infrastructure decisions arrive late or over budget.

Independent Practice

Founder & Fractional VP of AI

Through bhella.ai, founded in 2025, I work as a fractional VP of AI for organizations moving from pilot experimentation to governed production operations. I partner with executive teams to set technical vision, establish governance frameworks, and build the organizational structures that turn AI investments into production systems.

The work is hands-on. I design agentic platforms with multi-agent orchestration, structured tool use, and Model Context Protocol integration; architect LLMOps platforms spanning prompt management, dataset governance, and automated evaluation; and build retrieval systems whose quality is measured against a versioned test set rather than judged from a demo.

I hold my own work to that standard. Rasoi and Gurbani RAG, two reference retrieval systems I built and documented as companion books, record every design decision against an evaluation baseline — which is how Rasoi's retrieval hit rate went from 58% to 100%, and how a cross-encoder reranker lifted Gurbani RAG's ranking quality (MRR) by 32%.

I also stand up continuous evaluation infrastructure for agentic systems — trajectory grading, prompt and tool-call regression testing, and LLM-as-judge pipelines with human review — so silent model regressions are caught before they reach production, and I define governance frameworks aligned with the EU AI Act and SOC 2. Alongside the build work, I advise on AI investment strategy, build-vs-buy decisions, and organizational readiness, and guide engineering teams through the shift to AI-assisted development and agentic SDLC workflows.

Teaching & Mentorship

Adjunct Faculty — Computer Science

Since 2019, I've taught programming language design, systems architecture, and computational thinking at Clark College, reaching 100+ students annually using Python and Rust. I design project-based curriculum emphasizing real-world software engineering practices and mentor early-career engineers transitioning into professional platform engineering roles.