476774
R&D LLM Foundational Model Alignment
ENERGY, DEPARTMENT OF > ENERGY, DEPARTMENT OF > BROOKHAVEN NATL LAB -DOE CONTRACTOR
Archived
Quick analysis
DOE Brookhaven National Lab (BNL) Sources Sought for a 24‑month subcontract to conduct LLM–Foundation Model alignment within a neuroscience program (NeuroX‑Fusion).
SOW requires leadership in naturalistic fMRI and intracranial recordings, brain–language mapping, and LLM–brain encoding—specialized research capabilities not reflected in Trace’s core lanes or past performance.
BNL anticipates a noncompetitive acquisition unless additional qualified sources are identified, indicating a likely preferred research incumbent and a high barrier to entry.
No set-aside currently; NAICS 541715 with 1,000-employee size standard (Trace would qualify as small by headcount), but technical fit and competitive position remain poor.
Scope summary
- Design and validate LLM–Foundation Model alignment strategies and benchmarks for NeuroX‑MRI, NeuroX‑Ephys, and NeuroX‑Fusion.
- Co-design neuroscience-aligned pretraining/auxiliary objectives; deliver prototype code modules (PyTorch/HuggingFace) and training scripts with results notes.
- Produce benchmark specifications, evaluation scripts/notebooks, and integration/API requirements.
- Curate datasets subject to IRB/data-use agreements; provide preprocessed signals, time-aligned transcripts/annotations, and recommended splits.
- Monthly progress meetings; kickoff in first month; Year‑2 2–3 day hackathon including planning and outcomes report.
- Runnable examples on modest-scale models and public/lab brain datasets; emphasis on objective stability and reproducibility.
- Academic dissemination encouraged under DOE/BNL IP/publication policies.
Dimension scores
Total
- 21/100
Technical Fit
- 3/20
SOW demands LLM–brain alignment expertise, naturalistic fMRI/intracranial recordings, and neural coupling research (evidence: SOW requirements in ledger). Trace’s core capabilities center on SATCOM, CDS, MPE/coalition C2, tactical edge, SE&I—not neuroscience or LLM–neuro modeling.
Past Performance
- 1/20
No evidence of past performance with DOE/BNL or neuroscience/LLM‑brain alignment. Trace PP is DISA/DoD-centric (MPE-S, ABMS, SHIELD, GTACS II, etc.), which does not map to the NeuroX‑Fusion mission.
Vehicles / Contract Access
- 5/10
This is a Sources Sought (not on a vehicle). Full and open path possible, but BNL notes potential noncompetitive award unless additional sources are found; no Trace prime vehicle relevance indicated.
Competitive Position
- 1/15
BNL anticipates noncompetitive acquisition and seeks a leader in fMRI/iEEG and LLM–brain encoding—niche academic strengths Trace does not claim. This implies an incumbent/preferred research partner and Trace as an underdog.
Risk
- 4/15
High execution risk due to specialized neuroscience domain, IRB/data‑use constraints, and new customer/mission. Classification/logistics risks are low (SOW notes no export-controlled technical data), but capability and teaming risks are significant.
Strategic Value
- 1/10
Limited strategic alignment. Would open a DOE/BNL research relationship but outside Trace’s core lanes; minimal synergy with SATCOM/CDS/C5ISR focus.
Compliance
- 6/10
No set-aside; NAICS 541715 size standard 1,000 employees—Trace (~330) would qualify as small. No clearance/certification constraints cited. IRB/data-use practices may require teaming, but no hard compliance disqualifier is stated.
Concerns
- Specialized domain expertise required (fMRI/iEEG, LLM–brain alignment) not in Trace’s core capabilities.
- BNL anticipates noncompetitive award absent additional qualified sources, suggesting incumbent advantage.
- IRB and data-use constraints add process overhead and require established research governance.
- Unclear funding value and final acquisition approach at Sources Sought stage.
- New customer/mission area (DOE/BNL neuroscience) with limited transfer from Trace’s DoD communications pedigree.
Teaming opportunities
- Academic/research partner with demonstrated leadership in naturalistic fMRI and intracranial recordings.
- Experts in brain–language mapping and LLM–brain encoding with published benchmarks.
- IRB infrastructure and data governance for human neuroscience datasets.
- Research engineers proficient in PyTorch/HuggingFace for LLM/FM training and evaluation.
- Access to relevant brain datasets and computational resources tuned for multimodal neuro/LLM experiments.
Competitive position
- If pursued via teaming, position Trace as the secure systems integration and reproducibility backbone (pipelines, data governance, deployment hardening) while the research partner leads neuroscience and LLM–brain modeling.
- Leverage SE&I and integration strengths to package runnable, documented pipelines and evaluation harnesses aligned to SOW deliverables.
Bid/No bid factors
- Notice states BNL may conduct a noncompetitive acquisition unless additional sources are identified.
- SOW demands niche neuroscience expertise not reflected in Trace’s documented capabilities.
- Potential IRB/data-use burden without existing human-subject research infrastructure.
- Place of performance metadata lists Upton, KY while BNL is in Upton, NY—location discrepancy in notice metadata.
Documents
Download all
SOW_NeuroX_Fusion.pdf
Download
SOW_NeuroX_Fusion.pdf
Download
Summary
Statement of Work (dated 2026-03-09) for the DOE ASCR-funded NeuroX-Fusion project describing a 24-month subcontract to design and validate LLM–Foundation Model (FM) alignment for multimodal brain foundation models. Major workstreams cover (1) LLM–FM alignment design and benchmarks, (2) neuroscience-aligned pretraining/auxiliary objectives with prototype implementations, and (3) datasets, benchmarks, end-to-end prototype pipelines, and a Year‑2 hackathon; deliverables include design notes, benchmark specs and scripts, prototype code (PyTorch/HuggingFace-style), dataset packages (subject to IRB/data-use agreements), and publication reporting consistent with DOE/BNL IP policies.
- Document date: March 9, 2026; BNL procurement SOW for 'NeuroX-Fusion: LLM-Foundation Model Alignment'.
- Period of performance: Twenty-four (24) months from subcontract start date.
- Funding/owner: DOE ASCR-funded project executed by Brookhaven National Lab (BNL).
- Three major task areas: Task Area 1 — LLM–FM alignment (design, benchmarks, cross-modal validation); Task Area 2 — neuroscience-aligned pretraining objectives and prototype implementations; Task Area 3 — datasets, benchmarks/leaderboards, end-to-end pipelines, and a hackathon.
- Technical expertise required: naturalistic fMRI and intracranial recordings, brain–language mapping and LLM–brain encoding, neural coupling during storytelling/real-world communication.
- Deliverables include short design notes and playbooks (T1.1, T1.3), benchmark specs + example evaluation scripts/notebooks (T1.2, T3.2), prototype code modules (PyTorch/HuggingFace-style) and training scripts with results notes (T2.2), and integration/API requirements (T2.3).
- Implementation expectations: prototypes on public LLMs or moderate-scale variants and lab/public brain datasets; emphasis on runnable examples on modest-scale models and stability of objectives.
- Data handling constraints: dataset curation subject to IRB and data‑use agreements; deliverables must include preprocessed signals, time-aligned transcripts/annotations, and recommended train/val/test splits.
- Programmatic schedule: Year‑1 and Year‑2 milestone structure with monthly progress meetings, kickoff within first month, and a 2–3 day hackathon in Year‑2 (planning, execution, and outcome report deliverables).
- Academic dissemination: encouraged (conferences/journals) but subject to DOE and BNL policies on publications and intellectual property.
- Regulatory note in header: SOW 'contains no export-controlled technical data' and 'does not constitute development or production technology under applicable export control regulations.'
Technical details
Entry: sync
Review status: Up to date
Logical upstream opportunity: 6644
Notice lineage:
- e307b75073844e5dbb5eda97b10c92fd (posted 2026-03-30) - current
Last synced: 2026-03-31T11:26:05.711928+00:00
Last analyzed: 2026-03-30T23:08:51.577877+00:00
Latest package fingerprint: 044e1ba9cc826e794ed7f56d2e10008beec16a55639dd9469815162c979a1234
Latest package notice: e307b75073844e5dbb5eda97b10c92fd
Latest package documents: 1
Recent package history- 2026-03-30T23:07:21.858291+00:00: e307b75073844e5dbb5eda97b10c92fd with 1 docs
- 2026-03-30T18:31:03.526143+00:00: e307b75073844e5dbb5eda97b10c92fd with 1 docs
Labels
Human notes
No human notes yet.