I work with enterprise and product AI teams that need hands-on help on LLM post-training, agentic systems, or production retrieval and evaluation — not a slide-only advisory. Engagements are limited each quarter so the work stays deep. Typical work: architecture reviews, training-pipeline design, agentic system audits, and focused workshops. To start, send a short note.
A structured look at your LLM or agent stack: training and serving paths, eval gates, retrieval design, and where risk or cost concentrates. You leave with a written set of priorities and trade-offs, not a generic maturity score.
1–2 week review · async document + working session
Design or tighten SFT → preference/RL loops (including GRPO-style recipes), reward and data filters, and eval pre-flights so GPU spend tracks real uplift.
Multi-week design or co-build · clear milestones
Audit multi-agent or computer-use systems on finished-task cost, failure modes, tool/orchestration choices, and eval harnesses — beyond leaderboard scores.
1–3 week audit · findings + recommended experiments
Focused sessions for engineering or applied-research teams on post-training loops, agent eval, or production RAG failure modes. Practical, stack-aware, limited slides.
Half-day or full-day · remote or on-site by arrangement
Email or LinkedIn with context: team, stack, what “good” looks like, constraints (NDA, timeline, budget band if known).
30–45 minutes to test fit and outline options.
Scope, shape, timeline, and fee. Work starts only after that is agreed.
Prefer email for first contact — details on Contact.
Architecture reviews, training-pipeline design (SFT / preference / RL including GRPO-style loops), agentic system audits, and focused workshops. I do not take open-ended “be our fractional Head of AI” retainers by default.
Product and platform teams already building with LLMs or agents who need senior help on training, eval, orchestration, or production reliability. Early idea-stage pitches with no stack yet are usually a poor fit.
Send a short note via Contact. If there is a fit, we do a scoping call, then a written proposal before any paid work.
Yes, for engineering and applied-research groups — post-training, agent evaluation, or production RAG/eval. Scope and audience are fixed up front.
Based in Bengaluru (IST). Happy to overlap with US and EU teams for calls; deep work stays async-friendly.
Yes to NDAs when needed. Public write-ups are optional and only with approval; until then I will not invent or publish client stories on this site.
If this matches a problem on your team, get in touch. Include a few lines on the system, the pain, and the timeline.
Contact me