Service 01
AI Engineering Assessment
- Review of agent workflows, tests, and release controls
- Gaps between intended behavior and what gets checked
- Baseline for delivery time, rework, and defects
- Ranked risks and recommendations for leadership
Waypoint 01 // Engineering leadership
PWEP Consulting
I'm Sahil Talwar, former Oracle engineering director and RedShelf VP. I help engineering leaders adopt AI, improve delivery, and build stronger teams through independent assessments, practical plans, and leadership coaching.
50People led · Oracle
20%+Delivery gain via AI · RedShelf
~25%Lower operating costs · RedShelf
30BAPI calls / day · Oracle
Waypoint 02 // Engine Room
Where I can help
Independent advice grounded in the code, the workflow, and the people doing the work. I assess, recommend, and coach. Your team owns implementation.
Service 02
AI Adoption & Change Planning
- Where AI helps, where it does not, and why
- Pilot plans with success measures and stop criteria
- Recommendations for context, evaluations, and review gates
- Clear ownership and a staged adoption roadmap
Service 03
Engineering Organization Assessment
- Team interviews and a review of delivery evidence
- Bottlenecks in ownership, decisions, and handoffs
- Recommendations for team structure and management practices
- A prioritized improvement plan tied to business goals
Service 04
Leadership Coaching & Advisory
- Coaching for engineering managers and executives
- Technical strategy and architecture decision reviews
- Guidance through changes in tools, roles, and expectations
- Follow-up on outcomes as your team makes changes
When to bring me in
- AI produces more code, but review and rework keep growing.
- Nobody can show whether AI adoption improved delivery.
- Agents pass tests without meeting the actual requirement.
- Team structure and decision-making slow the roadmap.
Who I work with
- CTOs and VPs assessing AI adoption and delivery risk
- Founders rethinking engineering structure and priorities
- Engineering leaders navigating organizational change
Waypoint 03 // Casefile
Selected work
Across separate roles at Oracle and RedShelf, I led engineering for platforms supporting $185M in combined annual recurring revenue. Below: work from those roles and an independent AI project.
Job Hunter: evaluating agent workflows
Selected work- Built an open-source workflow to match jobs and tailor resumes against source material.
- In a single-job comparison, reduced runtime from over 7.5 hours to ~10 minutes with identical inputs and the same processing stages.
- Applied LLM-as-a-judge best practices, including blind grading and a separate truthfulness reviewer.
- Tested through a shared LLM interface: FakeLLM for unit tests, RealLLM for integration evaluations.
Oracle: operating at scale
Selected work- Led engineering for a $150M annual recurring revenue platform handling 30B API calls/day.
- Directed the IBM SoftLayer to Oracle Cloud migration and move from a monolith to services.
- Led the team that built a shared platform integrating four products from four acquired companies.
RedShelf: improving delivery
Selected work- Improved delivery velocity by 20%+ through AI-assisted engineering transformation, with human review and test gates.
- Led engineering for a platform supporting $35M+ in annual recurring revenue; gave managers shared delivery and quality metrics.
- Reduced legacy platform operating costs by ~25% and issue resolution time by ~30%.
Waypoint 04 // Protocol
How engagements run
- 01
Assess
Talk to the team, inspect the workflow and technical evidence, and establish a baseline. Separate observed problems from assumptions.
- 02
Recommend
Set out the options, tradeoffs, and priorities. Agree on owners, success measures, and what not to change.
- 03
Guide
Coach leaders through the changes and review decisions along the way. Your team owns implementation.
- 04
Review
Compare results with the baseline. Keep what works, address what does not, and leave the team able to continue without me.
Rules of engagement
- Control the workflow
- Models are probabilistic. State transitions, permissions, and release gates should not depend on model judgment alone.
- Show the evidence
- Define intended behavior before choosing checks. Measure delivery and defects, not just code volume or tool usage.
- Strengthen the team
- Make the reasoning clear, develop the people doing the work, and avoid creating a dependency on the consultant.
Formats
- Assess
- Focused review with written findings
- Plan
- Adoption roadmap and decision criteria
- Coach
- Support for engineering leaders and managers
- Advise
- Ongoing reviews of decisions and outcomes
Waypoint 05 // Contact
What needs to work?
Tell me where engineering is stuck, what you've tried, and what needs to change. Whether it's AI adoption, team structure, or a leadership decision, start with the problem.
Beyond the work // Sahil Talwar
The person behind the practice: what I build, what I read, and the questions I keep coming back to.
Meet Sahil Talwar© 2026 PWEP Consulting · Washington, DC · Built with Astro + Three.js