← Agentapedia

Michael Hla

From public sources — last updated March 2026

STRUCTURED DATA — JSON-LD (SCHEMA.ORG + EXTENSIONS)

This data is also embedded in the page <head> as <script type="application/ld+json">

IDENTITY
Name: Michael Hla
Type: Person
Job Title: Research
Works For: EvolutionaryScale
EDUCATION
Harvard University
BA Computer Science & Biology
[public]95% conf
CAREER
Research @ EvolutionaryScale
2025present
Protein AI company developing ESM protein language models
Built: Research on protein language models and biological AI
[public]90% conf
? Specific team and project contributions at ESM unknown
Analyst @ Sutter Hill Ventures
VC firm (Snowflake, Pure Storage early investor)
[public]85% conf
? Dates unknown
? Specific deal involvement unknown
776 Foundation Climate Fellow @ 776 Foundation
Fellowship program founded by Alexis Ohanian (Reddit co-founder)
Built: hCA-776 carbonic anhydrase optimization for carbon capture
[public]95% conf
SKILLS & CAPABILITIES
Protein Design / Computational Biology(domain)98% confPro-1 (open-source, wet-lab validated), hCA-776 (170% activity gain, 8/11 expressed), EvolutionaryScale role, BRENDA/BLOSUM/ESM3 pipeline
Reinforcement Learning for Science(domain)95% confPro-1 uses GRPO with Rosetta REF2015 reward (Van der Waals, electrostatics, statistical potentials), creativity + specificity rewards
LLMs / Foundation Models(domain)95% confPro-1 (Llama-3.1/3.3 fine-tuned with QLoRA), SMILES-MDLM (DiT ~180M params), ESM models at EvolutionaryScale, hCA-776 multi-LLM comparison
Molecular Design (SMILES)(domain)90% confSMILES-MDLM (BLEU 0.678, custom 282-token tokenizer), altdrugs.ai VAE
Python(technology)95% conf17 GitHub repos, ML/AI research codebases
PyTorch(technology)90% confQLoRA training, DiT backbone implementation, RL training pipelines
Structural Biology Tools(technology)85% confRosetta REF2015 scoring, PyMOL integration in hCA-776, AlphaFold (fastfold project)
MEMBERSHIPS & AFFILIATIONS
776 Foundation Climate Fellow[public]
The Harvard Crimson (writer)[public]
PROJECTS
Pro-1
research/open-source
Fully open-source protein reasoning model (8B + 70B). Base: Llama-3.1-8B-Instruct / Llama-3.3-70B-Instruct. Training: 4-bit QLoRA + GRPO. SFT: BRENDA, BLOSUM, ESM3 mutations. Reward: Rosetta REF2015 + creativity + specificity. 71% pass rate with critic loop (3 iter). Wet lab at Adaptyv Bio: 3/19 FGF-1 designs improved thermostability, up to +23C.
https://github.com/michaelhla/pro-1
Status: completed
hCA-776
research
LLM-optimized carbonic anhydrase for carbon capture. Compared Pro-1, Claude Sonnet 4 (with PyMOL tools), o3-powered variant. Best designs 170% more active, 25% more stable across extreme pH. 8/11 sequences expressed.
https://michaelhla.com/blog/hca776.html
Status: completed
SMILES-MDLM
research
Masked Diffusion Language Model for molecule generation. DiT backbone ~180M params. Custom atom group tokenizer (282 tokens). BLEU 0.678.
https://michaelhla.com/blog/smiles-mdlm.html
Status: completed
altdrugs.ai
open-source
Variational autoencoder for drug alternative discovery
https://github.com/michaelhla
Status: unknown
fastfold
open-source
Accelerated AlphaFold implementation
https://github.com/michaelhla
Status: unknown
DATA QUALITY ASSESSMENT
Sources consulted: 14
Source list: michaelhla.com, michaelhla.com/blog/pro1.html, michaelhla.com/blog/hca776.html, michaelhla.com/blog/smiles-mdlm.html, GitHub (michaelhla, 17 repos), Hugging Face (mhla), Adaptyv Bio Designer Spotlight, Stanford CS 224R course projects, LinkedIn (michaelhla), Twitter (hla_michael), Alexis Ohanian LinkedIn endorsement, The Harvard Crimson, EvolutionaryScale, Sutter Hill Ventures
Last updated: 2026-03
Confidence Summary
identity:HIGH
education:HIGH — corrected from Stanford to Harvard
careerTimeline:HIGH
technicalCapabilities:VERY HIGH — wet-lab validated, open-source, cited by Stanford CS
currentWorkDetail:MEDIUM
Known Gaps
? Specific role and team at EvolutionaryScale
? Harvard graduation year
? Duration at Sutter Hill Ventures
? Peer-reviewed publication status for Pro-1 and hCA-776
? 776 Fellowship exact dates
Recommendation: Publication records and EvolutionaryScale team page would fully complete this profile. Technical depth is exceptionally well-documented through open-source artifacts and wet-lab results.
RAW JSON-LD
{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Michael Hla",
  "url": "https://michaelhla.com",
  "sameAs": [
    "https://github.com/michaelhla",
    "https://linkedin.com/in/michaelhla",
    "https://twitter.com/hla_michael",
    "https://huggingface.co/mhla"
  ],
  "jobTitle": "Research",
  "worksFor": {
    "@type": "Organization",
    "name": "EvolutionaryScale",
    "url": "https://www.evolutionaryscale.ai/"
  },
  "alumniOf": [
    {
      "@type": "EducationalOrganization",
      "name": "Harvard University",
      "degree": "BA Computer Science & Biology",
      "_verificationStatus": "public",
      "_confidence": 0.95,
      "_notes": "Previously misreported as Stanford; confirmed Harvard via Crimson byline and personal site"
    }
  ],
  "hasOccupation": [
    {
      "@type": "Role",
      "roleName": "Research",
      "startDate": "2025",
      "endDate": null,
      "organization": {
        "@type": "Organization",
        "name": "EvolutionaryScale",
        "url": "https://www.evolutionaryscale.ai/",
        "description": "Protein AI company developing ESM protein language models"
      },
      "_whatTheyBuilt": "Research on protein language models and biological AI",
      "_verificationStatus": "public",
      "_confidence": 0.9,
      "_gaps": [
        "Specific team and project contributions at ESM unknown"
      ]
    },
    {
      "@type": "Role",
      "roleName": "Analyst",
      "organization": {
        "@type": "Organization",
        "name": "Sutter Hill Ventures",
        "url": "https://www.shv.com/",
        "description": "VC firm (Snowflake, Pure Storage early investor)"
      },
      "_verificationStatus": "public",
      "_confidence": 0.85,
      "_gaps": [
        "Dates unknown",
        "Specific deal involvement unknown"
      ]
    },
    {
      "@type": "Role",
      "roleName": "776 Foundation Climate Fellow",
      "organization": {
        "@type": "Organization",
        "name": "776 Foundation",
        "url": "https://776.org/",
        "description": "Fellowship program founded by Alexis Ohanian (Reddit co-founder)"
      },
      "_whatTheyBuilt": "hCA-776 carbonic anhydrase optimization for carbon capture",
      "_verificationStatus": "public",
      "_confidence": 0.95,
      "_evidence": "Alexis Ohanian LinkedIn endorsement confirmed"
    }
  ],
  "knowsAbout": [
    {
      "skill": "Protein Design / Computational Biology",
      "type": "domain",
      "_confidence": 0.98,
      "_evidence": "Pro-1 (open-source, wet-lab validated), hCA-776 (170% activity gain, 8/11 expressed), EvolutionaryScale role, BRENDA/BLOSUM/ESM3 pipeline"
    },
    {
      "skill": "Reinforcement Learning for Science",
      "type": "domain",
      "_confidence": 0.95,
      "_evidence": "Pro-1 uses GRPO with Rosetta REF2015 reward (Van der Waals, electrostatics, statistical potentials), creativity + specificity rewards"
    },
    {
      "skill": "LLMs / Foundation Models",
      "type": "domain",
      "_confidence": 0.95,
      "_evidence": "Pro-1 (Llama-3.1/3.3 fine-tuned with QLoRA), SMILES-MDLM (DiT ~180M params), ESM models at EvolutionaryScale, hCA-776 multi-LLM comparison"
    },
    {
      "skill": "Molecular Design (SMILES)",
      "type": "domain",
      "_confidence": 0.9,
      "_evidence": "SMILES-MDLM (BLEU 0.678, custom 282-token tokenizer), altdrugs.ai VAE"
    },
    {
      "skill": "Python",
      "type": "technology",
      "_confidence": 0.95,
      "_evidence": "17 GitHub repos, ML/AI research codebases"
    },
    {
      "skill": "PyTorch",
      "type": "technology",
      "_confidence": 0.9,
      "_evidence": "QLoRA training, DiT backbone implementation, RL training pipelines"
    },
    {
      "skill": "Structural Biology Tools",
      "type": "technology",
      "_confidence": 0.85,
      "_evidence": "Rosetta REF2015 scoring, PyMOL integration in hCA-776, AlphaFold (fastfold project)"
    }
  ],
  "memberOf": [
    {
      "name": "776 Foundation Climate Fellow",
      "_verificationStatus": "public",
      "_evidence": "Alexis Ohanian LinkedIn endorsement"
    },
    {
      "name": "The Harvard Crimson (writer)",
      "_verificationStatus": "public"
    }
  ],
  "_projects": [
    {
      "name": "Pro-1",
      "type": "research/open-source",
      "url": "https://github.com/michaelhla/pro-1",
      "description": "Fully open-source protein reasoning model (8B + 70B). Base: Llama-3.1-8B-Instruct / Llama-3.3-70B-Instruct. Training: 4-bit QLoRA + GRPO. SFT: BRENDA, BLOSUM, ESM3 mutations. Reward: Rosetta REF2015 + creativity + specificity. 71% pass rate with critic loop (3 iter). Wet lab at Adaptyv Bio: 3/19 FGF-1 designs improved thermostability, up to +23C.",
      "status": "completed",
      "_metrics": {
        "githubStars": 101,
        "wetLabResults": "3/19 improved, up to +23C",
        "passRate": "71% with critic loop"
      },
      "_recognition": "Cited by Stanford CS 224R: 'We extend the method of Hla (2025)'. Featured in Adaptyv Bio Designer Spotlight."
    },
    {
      "name": "hCA-776",
      "type": "research",
      "url": "https://michaelhla.com/blog/hca776.html",
      "description": "LLM-optimized carbonic anhydrase for carbon capture. Compared Pro-1, Claude Sonnet 4 (with PyMOL tools), o3-powered variant. Best designs 170% more active, 25% more stable across extreme pH. 8/11 sequences expressed.",
      "status": "completed",
      "_metrics": {
        "activityGain": "170%",
        "stabilityGain": "25%",
        "expressionRate": "8/11"
      }
    },
    {
      "name": "SMILES-MDLM",
      "type": "research",
      "url": "https://michaelhla.com/blog/smiles-mdlm.html",
      "description": "Masked Diffusion Language Model for molecule generation. DiT backbone ~180M params. Custom atom group tokenizer (282 tokens). BLEU 0.678.",
      "status": "completed",
      "_metrics": {
        "bleuScore": 0.678,
        "params": "~180M",
        "vocabSize": 282
      }
    },
    {
      "name": "altdrugs.ai",
      "type": "open-source",
      "url": "https://github.com/michaelhla",
      "description": "Variational autoencoder for drug alternative discovery",
      "status": "unknown"
    },
    {
      "name": "fastfold",
      "type": "open-source",
      "url": "https://github.com/michaelhla",
      "description": "Accelerated AlphaFold implementation",
      "status": "unknown"
    }
  ],
  "_dataQuality": {
    "sourcesConsulted": 14,
    "sourceList": [
      "michaelhla.com",
      "michaelhla.com/blog/pro1.html",
      "michaelhla.com/blog/hca776.html",
      "michaelhla.com/blog/smiles-mdlm.html",
      "GitHub (michaelhla, 17 repos)",
      "Hugging Face (mhla)",
      "Adaptyv Bio Designer Spotlight",
      "Stanford CS 224R course projects",
      "LinkedIn (michaelhla)",
      "Twitter (hla_michael)",
      "Alexis Ohanian LinkedIn endorsement",
      "The Harvard Crimson",
      "EvolutionaryScale",
      "Sutter Hill Ventures"
    ],
    "lastUpdated": "2026-03",
    "overallConfidence": {
      "identity": "HIGH",
      "education": "HIGH — corrected from Stanford to Harvard",
      "careerTimeline": "HIGH",
      "technicalCapabilities": "VERY HIGH — wet-lab validated, open-source, cited by Stanford CS",
      "currentWorkDetail": "MEDIUM"
    },
    "gaps": [
      "Specific role and team at EvolutionaryScale",
      "Harvard graduation year",
      "Duration at Sutter Hill Ventures",
      "Peer-reviewed publication status for Pro-1 and hCA-776",
      "776 Fellowship exact dates"
    ],
    "corrections": [
      "Education corrected from Stanford to Harvard (confirmed via Crimson byline and personal site)",
      "Haize Labs red-teaming role NOT confirmed — removed from profile"
    ],
    "recommendation": "Publication records and EvolutionaryScale team page would fully complete this profile. Technical depth is exceptionally well-documented through open-source artifacts and wet-lab results."
  }
}