Credit
The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.
What they publicly do
Assistant Professor at MIT EECS and CSAIL who leads the MIT OASYS lab; author of ColBERT and DSPy, co-author of GEPA and Recursive Language Models (RLMs) [Verified omarkhattab.com]. Teaches a new MIT subject, 6.S978 AI Engineering, in Fall 2026 [Verified omarkhattab.com].
- X profile: x.com/lateinteraction
- GitHub: github.com/okhat
- Site: omarkhattab.com
- Blog: github.com/okhat/blog
- Also: dspy.ai
- Also: ColBERT.ai
- Also: mit-oasys.github.io
- Also: gepa-ai.github.io/gepa
Public work
- DSPy repo MIT last activity 2026-09-25 · 38,265 GitHub stars on 2026-09-24
Declarative framework for programming, not prompting, LM pipelines: signatures, modules, optimizers, and an evaluation API that its own CI exercises [Verified repo; franken-research stack/agent-frameworks.md:20].
- GEPA repo MIT last activity 2026-09-24 · 6,731 GitHub stars on 2026-09-24
Genetic-Pareto optimizer: LLM reflection over execution traces plus Pareto-efficient evolutionary search to optimize any text parameter (prompts, code, agent architectures, skills) against a metric; paper is an ICLR 2026 oral [Verified repo README; arxiv.org/abs/2507.19457].
- Recursive Language Models (co-author) repo MIT last activity 2026-09-24 · 5,646 GitHub stars on 2026-09-24
Inference paradigm where the prompt lives as a variable in a REPL and the root LM recursively calls sub-LMs over slices of it; NeurIPS 2026 [Verified omarkhattab.com, repo README].
- ColBERT repo MIT last activity 2025-10-14 · 3,944 GitHub stars on 2026-09-24
Late-interaction neural retrieval (token-level MaxSim), with PLAID engine [Verified repo; omarkhattab.com].
- WARP: An Efficient Engine for Multi-Vector Retrieval paper no license stated last activity 2025
Multi-vector retrieval engine; SIGIR 2025 Best Paper [Verified omarkhattab.com].
- Meta-Harness: End-to-End Optimization of Model Harnesses paper no license stated last activity 2026
Optimizes the harness around a model end to end; COLM 2026 [Verified omarkhattab.com; contents not read].
Licenses of GitHub repositories are from the authenticated GitHub license API on 2026-09-25, one row per repository in evidence/licenses.jsonl (“no license file” means the API found none; “license file present, not recognised by GitHub” means GitHub could not classify it). Other licenses are as the source states. Check the license file before reusing anything.
Relevance to our work
5 of 5 lenses: agents, evals, search, research-method, orchestration
DSPy and GEPA are MIT-licensed [Verified: GitHub license API, 2026-09-25, evidence/licenses.jsonl]. GEPA optimizes a text artifact, such as a skill file or a judge prompt, against scored tasks, and gskill applies it to skills for coding agents [Verified: gepa-ai.github.io/gepa/blog/automatically-learning-skills-for-coding-agents].
Where it already touches our work: Franken Research already names DSPy in its agent-stack verdicts.
Deep dive:
- Omar Khattab and alex zhang, GEPA for skills and Recursive Language Models for long context (written 2026-09-25)
Scored 0 to 5 against the study’s relevance rubric: 5 means public, openly licensed work squarely in our lanes that we could study or adopt now, and 1 means little connection to our work. The score is one reviewer’s judgement of how much this public work bears on our own work. It is not a ranking of people and says nothing about the person or the quality of their work.
What we would study next
- GEPA 'automatically learning skills for coding agents' and optimize_anything (gepa-ai.github.io/gepa/blog/automatically-learning-skills-for-coding-agents). The published method for optimizing a skill file against scored tasks, with a sealed test set. Effort: medium.
- Meta-Harness paper (arxiv.org/abs/2603.28052). Optimizes the agent harness itself; not read for this study. Effort: small.
- WARP / ColBERT late-interaction engine (arxiv.org/abs/2501.17788). The reference design for late-interaction retrieval. Effort: medium.
Before reusing anything: DSPy, GEPA, ColBERT and rlm are MIT per the GitHub license API (2026-09-25). DSPy vendors LiteLLM MIT text under dspy/clients/model_metadata (already recorded in franken-research stack/licenses.tsv:87). ColBERT model checkpoints on Hugging Face were not license-checked.
Sources
- omarkhattab.com
- github.com/stanfordnlp/dspy
- github.com/gepa-ai/gepa
- github.com/stanford-futuredata/ColBERT
- github.com/alexzhang13/rlm
- arxiv.org/abs/2507.19457
- arxiv.org/abs/2603.28052
- arxiv.org/abs/2501.17788
Overall confidence in this record: High. Sources were read on 2026-09-24 and every link was checked on 2026-09-25. The full record is line 11 of people.jsonl.
Limits
Not verified: Download figures ('several millions of times every month') are self-reported. GEPA README results table (e.g. '90x cheaper', '55% to 82%' on Jinja) not reproduced. Meta-Harness and WARP contents not read beyond titles/venues.
This record comes from a first-pass survey of public material by an AI agent session on 2026-09-24, edited for publication on 2026-09-25. Nothing was installed or run for it. Roles, stars and licenses change; see the study’s limits.
Request a correction
Is something about this record wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.