Dhruv Bhatia is a founding researcher at Morph, working on specialized models and inference for coding agents. His published work includes accounts of fast code-edit application and model-assisted code search.[1][2][3]
Start with Fast Apply Makes Faster Agents →
Links
Morph researchFounding researcher · Morph
Contributions
Build the inference system around the task
In Morph’s AWS case study, Bhatia describes engineering a custom inference engine and GPU scheduling layer for stable code-merging performance under concurrent demand. Fast Apply uses a specialized model to incorporate proposed edits into existing files. The work focuses on the infrastructure around an agent’s reasoning rather than treating token generation alone as the product. Reported speed and customer benefits are company case-study results, not independent measurements.[1]
Give code search its own model
Bhatia co-authored Morph’s January 2026 WarpGrep introduction with Dat Quoc and Tejas Bhakta. It presents reinforcement learning and parallel retrieval as ways to build a model specialized for finding code context. This complements edit application: a coding agent needs relevant files before it can decide what to change.[3]
Tools, writing & talks
Writing, research & talks
Start here
Fast Apply Makes Faster Agentsmorphllm.comA joint account of specialized code-edit models and the infrastructure serving them.
Begin with the distinction between proposing a change and applying it to a file. Software-development experience helps; compare correctness and latency on your own workload rather than relying only on headline throughput.
Co-author[2]- WarpGrep: Fast, Parallel Code Retrieval with RLmorphllm.com
How Morph approaches code-context retrieval with a specialized search model.
Co-author[3]