Featured
Ferrum Platform
17+ reposComposable platform for autonomous software engineering agents: execution, memory, evals, local inference, gateway, and tooling.
End-to-end systems for autonomous coding agents.
Staff-level AI infrastructure engineer — Rust, Python, LLM systems, agent memory, evals, and continual learning.
Calgary, Canada🇨🇦 Canadian citizenOpen to remote
SubstackMediumLinkedInScholarX
Now — Founding/sole engineer at General Intelligence building a synthetic-data marketplace; shipping the Ferrum agent-infra stack in open source alongside.
6 projects
Featured
Composable platform for autonomous software engineering agents: execution, memory, evals, local inference, gateway, and tooling.
Featured
Continual-learning substrate for coding agents — episode capture, replay, skill distillation, sandboxed SOAR search, and LoRA SFT.
Dynamic batching, SSE streaming, three-tier Metal/CUDA/CPU fallback, TurboQuant KV-cache compression.
Policy-checked commands, workspace snapshots, capped output/artifacts, replayable results. Firecracker-ready design.
HTTP, LLM, MCP, and agent traffic with routing, auth, limits, streaming, observability, token budgets, and tool policy.
Canonical mandates, SD-JWT/dSD-JWT chains, JOSE crypto, A2A helpers, constraint verification, reference actors, CLI demos.
Pinned six
What modern RL post-training implies for the next generation of AI.
Implementing provably near-optimal 7× KV-cache compression inside a Rust inference server.
How production AI systems orchestrate retrieval, memory, and context windows.
A field-tested synthesis of RAG, agents, and the PoC-to-production gap.
Why protocol-first (MCP, A2A) wins over framework lock-in.
Research survey across 600+ AI papers: continual RL benchmarking, plasticity crisis, GVFs as proto-world-models.
Paper · Apr 2026
Formalizes episode capture, replay, rule-based distillation, and evaluation; baseline-vs.-integrated protocol separates substrate learning from native agent memory.
Open →Architecture · Apr 2026
A composable platform for autonomous SWE: coding agents execute TDD loops, call typed tools, log episodes, replay failures, and measure continual-learning gains across sessions.
Open →Guide · 2023
Exploratory guide to prompt engineering techniques and patterns for working with large language models.
Open →AMS · 2017
Reiche, Robinson, Kay, Craun, Goswami, Bass. Wind shear climatology for terminal-radar approach airspace at seven airports.
Open →Aug 2024 — Present
Calgary, AB
Sole engineer building a two-sided synthetic-data marketplace for AI model training and fine-tuning, from zero to production-deployable.
Sep 2023 — Aug 2024
Calgary, AB
First engineering hire. Scaled the team 1→5, set the technical roadmap, established documentation and code-review practices, ran hiring, and drove ISO 27001 certification engineering changes.
Nov 2021 — Feb 2023
Burlington, ON
Worked in a 6-person pod supporting high-value clients on a B2B wealth-management platform reaching 22M+ end customers, including a $13B AUM firm.
Oct 2017 — Oct 2019
India
First engineering hire after the co-founders. Helped recruit, onboard, and technically mentor the team as the company scaled 1→35 engineers across its blockchain security practice.
Jul 2015 — Oct 2017
Reston, VA
Built weather analysis systems for aviation — accuracy directly affected air-traffic safety.