Head of Engineering · Concierto.cloud (Trianz)
Building AI-native platforms - and the teams that ship them.
I'm Gaurav Mittal. I lead 200+ engineers building enterprise SaaS for 80+ Fortune 500 clients. I designed data connector IP acquired by AWS, and I build production agentic AI - from fine-tuning LLMs on H100s to autonomous agents that take a bug report to a merged PR in under 20 minutes. IIM Ahmedabad alumnus, 20+ years in the craft.
200+
Engineers led
80+
Fortune 500 clients
99.9%
Platform uptime
20+
Years building
Flagship engineering
Systems I've built and shipped
Acquired by AWS
Data connector IP, now in AWS Athena
- Federation connectors for Snowflake, BigQuery, IBM DB2, and Oracle - acquired by AWS and now integrated into AWS Athena.
- Collaborated directly with AWS engineering on the code transition.
- Built on deep Trino internals work: custom connectors, query optimizations, and federation for enterprise-scale cross-source querying.
- Currently engaged with the AWS CloudWatch team building security connectors.
Live servers, near-zero downtime
CMigrate, a Rust block-level replication engine
- Three-crate Rust workspace (source agent, receiver, wire protocol) built on async Tokio.
- The agent streams disk blocks from live Linux/Windows servers over dual TLS connections, with lz4/zstd compression and xxHash checksums.
- Full sync, then checksum-based incremental scans, then a quiesced multi-disk cutover - with durability barriers and resume state on both ends.
- Ships as signed static binaries for Linux and Windows, with a Python/FastAPI control plane and React dashboard around it.
Bug → merged PR in <20 min
Maestro, a multi-agent bug-fix orchestrator
- Multi-agent pipeline: an analyser reproduces and localizes the defect, a fixer proposes the change, a validator runs the test suite, and a git agent shepherds the branch.
- Human approval gates on both the diagnosis and the final diff - engineers keep the judgment, agents do the legwork.
- Workspace isolation: every fix attempt runs in its own sandbox, so a wrong hypothesis never touches shared state.
- Result: bug report to merged PR in under 20 minutes for a meaningful class of bugs.
Hours → minutes
Concierto Sentinel, autonomous PR review
- Reviews every raised pull request end-to-end: code quality, architectural standards, security, and test coverage.
- Leaves inline comments and issues approve / request-changes verdicts with zero human kickoff.
- Paired with multi-agent automated security reviews: vulnerability scanning, dependency audits, and compliance checks.
- Manual review cycles reduced 60%+; review turnaround cut from hours to minutes.
96% accuracy · <800ms
Enterprise RAG at production scale
- Hybrid search: dense embeddings (OpenAI ada-002 plus BGE models fine-tuned on domain data) fused with sparse BM25 retrieval via Elasticsearch.
- Cross-encoder reranking (ms-marco-MiniLM) lifted top-5 retrieval precision from 72% to 94%.
- End-to-end latency cut from 4.2s to sub-800ms through Redis embedding caching, async chunk prefetching, and dynamic context-window packing.
- LLM evaluation harness holding 96% answer accuracy across 50K+ enterprise documents.
Zero external calls
Private AI coding CLI on self-hosted Gemma 3
- Terminal-native developer assistant powered by self-hosted Gemma 3 running on in-house GPUs.
- Codebase-aware context retrieval, tool use, and multi-step task execution: read, edit, run, verify.
- Zero external model calls - fully private AI-assisted development inside the enterprise security perimeter.
- Built on prior LLM work: fine-tuned open-source models on H100 GPUs (95% accuracy for internal knowledge search) with a deployment pipeline on GCP GPU clusters.
2–5X velocity
AI-native SDLC across the organization
- AI agents integrated across the full lifecycle: requirements, architecture, coding, testing, and deployment.
- Measured 2–5X velocity improvement across the engineering organization.
- 6+ production business agents - FinOps, Claims Processing, Knowledge AI, Migration Accelerator - running multi-step reasoning with self-reflective loops at 95% accuracy.
- Production agents reducing operational costs by 40%+ with measurable ROI.
Key achievements
The track record
- Scaled engineering from 40 to 200+ across five product clusters - attrition down 30%, time-to-productivity halved
- Built the SaaS platform and data IP that contributed to Accenture’s acquisition of Core Compete
- Open-source contributor: Trino, AWS Athena federation framework, Polaris Catalog
- 10M+ daily transactions at 99.9% uptime; post-release defects down 60% through shift-left engineering
Writing
Notes on engineering leadership & agentic AI
- Aug 16, 2026From Bug Report to Merged PR in Under 20 Minutes
How we built Maestro, a multi-agent bug-fix orchestrator with human approval gates - and what it changed about how our engineers work.
- Aug 10, 2026Scaling from 40 to 200 Engineers Without Losing the Plot
Value streams, end-to-end ownership, and the org-design choices that cut attrition by 30% while we 5X'd the team.