αf
System architecture

From raw data to
structured conviction

Four layers, each building on the last. Filings, fundamentals, and news become research you can act on — with every claim traceable to its source.

01
Data
02
Context Engine
03
Agent Systems
04
Research Output
αf
ace
Context Engine
STEP 01
Data ingestion
Primary sources pulled automatically and organized into a four-tier trust hierarchy. Every data point is versioned and traceable to its origin.
Regulatory filings
10-K, 10-Q, 8-K, S-1, and proxy statements — the authoritative corporate disclosure stream.
Daily Sub-minute
Fundamentals
Point-in-time income statements, balance sheets, cash flows, ratios, and analyst estimates.
Daily Structured
Earnings transcripts
Full quarterly calls with guidance extraction, tone scoring, and Q&A analysis.
Quarterly <24h lag
News & event flow
Market news routed, deduplicated, and scored for thesis relevance across every covered name.
Daily Multi-source
Macro indicators
Rates, employment, CPI, yield curves — the macro backdrop every thesis has to survive.
Daily Official series
Market & technicals
Equity pricing, options flow, volatility surfaces, and cross-sectional momentum.
Daily Tick-level
Daily ingest
ace://intake
Unified intake
Every endpoint writes into the context store once per cycle with source attribution, timestamp, and tier.
[Apr 15 · 00:05] Filings · overnight sync → 12 documents [Apr 15 · 00:08] Fundamentals · refresh → 49 entities [Apr 15 · 00:11] Macro · series update → 10Y yield 4.32% [Apr 15 · 00:14] News · 247 items routed → deduplicated [Apr 15 · 00:18] Market · close prices → persisted [Apr 15 · 00:21] ace · brief generated → ready
34+
Endpoints
5
Source types
4
Trust tiers
24/7
Refresh
STEP 02
Alpha Context Engine · ace
The structured foundation every agent reads from and writes to. Ground truth meets LLM reasoning, with every field traceable back to its tier.
Company records
Versioned ground-truth entities. Every agent reads from the same authoritative record; every write is audited.
49 companies12 snapshots/day
Knowledge graph
Entities linked across supply chain, competitive, and thematic dimensions. Queryable and propagating.
325 edges7 categories
Source hierarchy
Four tiers from primary filings to LLM inference. Every data point carries its tier so agents weight accordingly.
4 tiersWeighted
Filings
T1
Fundamentals
T1
Macro
T1
News
T3
Inference
T4
Market
T2
ace Context Engine
STEP 03
Multi-agent systems
Six specialized agents running in parallel around the context engine. Each reads what it needs, performs its task, and writes results back — no drift, no silos.
ace
Context Engine
READ · WRITE
News
Ingests, deduplicates, and routes actionable market signals.
ACTIVE · 247/hr
Earnings
Synthesizes beat/miss, guidance delta, and management tone.
ACTIVE · Q4 FY26
Monitor
Watches thesis conditions and flags pivot variables daily.
ACTIVE · 19 pivots
Decision
Gates relevance and routes actions via ActionManifest.
ACTIVE · 360 changes
Sentiment
Decay-weighted scoring normalized across coverage.
ACTIVE · NFSI 0.87
Temporal
Bayesian scenario inference across multiple horizons.
ACTIVE · Bull 89%
STEP 04
Research output - Sample
Not data dumps. Not summaries. Structured, traceable research your team can review, challenge, and build on — delivered on a cadence you control.
MRVL — Valuation & Scenarios
Idea Brief
Peer comparison
TickerRevenueEV/EBITDAFwd P/E
MRVL$8.19B17.4×~27.9×
AVGO$56B22.0×~24.0×
AMD$26B22.0×~22.0×
BULL Q1 earnings outperformance drives re-rating $186.31 +45%
BASE Consensus path holds, concentration contained $147.76 +15%
BEAR Loss of Amazon or Microsoft ASIC programs $96.37 −25%
+39.4%
EV-weighted return · target $179.17
9.6×
EV / Rev
17.4×
EV / EBITDA
27.9×
P/E (TTM)
0.7×
PEG (Fwd)
MRVL — Q4 FY2026 Earnings
Mar 5, 2026
Revenue accelerated +22% YoY in Q4
Gross margin 51.7% · EBITDA margin 32.9% · FCF $0.259B
MetricValuePriorΔ
Revenue$2.22B$1.82B+22% YOY
Gross margin51.7%—Reported
EPS (non-GAAP)$0.80$0.60+33% YOY
FCF$0.259B—Positive
Earnings verdict
THESIS INTACT
Growth: Confirming · Margins: Expanding · Guidance: Strong Beat-and-Raise
Pivot variables post-earnings
FY2027 Q1 earnings release (~May 2026)Pending
FY2027 full-year revenue ramp to ~$11BPending
20+ custom AI chip tape-outs to volumePending
MRVL — Sentiment & Signal
Daily update
Market signal: constructive
0.87
−1 Bearish0+1 Bullish
Peak score · Apr 1
7
Articles in window
3/7
Confirms bull
0.87
Daily score
0.98
7-day avg
Rolling 30-day sentiment
Mar 14 Mar 28 Apr 7
Signal interpretation
Market sentiment is constructive (NFSI 0.87). 7 articles in the latest day; 3 confirm bull vs 0 bear. Current score below 7-day average (0.98).
MRVL — Conviction Scorecard
LONG
EV-weighted return+39.4%
Revenue growth (FY2026)+42.1%
Gross margin51%
EBITDA margin32.1%
FCF margin17.1%
Actionability score0.60
Confidence score0.63
Customer/StrategicConcentration risk
Technical setupbullish · accumulation
Scenario probabilities
BULL
89% $186.31
BASE
6% $147.76
BEAR
6% $96.37
+39.4%
EV-weighted return
HIGH 4/5
Conviction · LONG
89%
Bull probability
MRVL — Investment Conclusion
LONG
High-conviction long on MRVL — asymmetric upside.
LONG-rated with +39.4% EV-weighted return (target $179.17). Marvell Technology is positioned as one of the most direct beneficiaries of hyperscale AI infrastructure buildout, with its custom ASIC and optical DSP franchises generating revenue acceleration strong enough to support a path from roughly $8B in FY2026 to $11B in FY2027 and $15B in FY2028 — a 34%+ compounding growth trajectory validated by 20+ custom AI chip tape-outs ramping toward volume production across CY2026–2027.

The signal stack is unambiguously constructive: near-term fund sentiment (NFSI bullish at 0.99), strong cluster momentum (0.80), margin expansion acceleration (0.49), and a macro tailwind supporting data center capex all reinforce the fundamental case. The primary risk — loss of the Amazon Trainium or Microsoft Maia ASIC programs — is real given top-four hyperscaler concentration, but the breadth of 20+ tape-outs and the secular irreversibility of custom AI silicon demand materially limits the downside scenario to a contained repricing rather than a thesis break.
Pivot variables monitored 19 active · Mixed states
FY2027 Q1 earnings release (~May 2026) with $2.4B revenue guidancePending
FY2027 full-year revenue ramp to ~$11B guidedPending
20+ custom AI chip tape-outs ramping to volume productionPending
Amazon Trainium program partner shift (risk/catalyst)Pending
MRVL — Initiation of Coverage
LONG
Initiation of coverage
MRVL
Marvell Technology, Inc.
Fabless semiconductor company designing custom AI processors, optical DSPs, switching ICs, and storage controllers primarily for hyperscale data centers.
ExchangeNASDAQ
SectorInformation Technology
Market Cap$76.86B
Coverage initiatedMarch 26, 2026
M
MRVL
Marvell Technology, Inc.
+39.4%
EV-weighted return
$8.2B
FY2026 Revenue
+42.1% YoY
51.0%
Gross Margin
+970bps
$2.6B
EBITDA
+304.6%
$1.4B
Free Cash Flow
+0.7% YoY
Coverage universe
8 of 49 companies
BA
BAC
LONG
Bank of America Corp.
+24.4%
EV-weighted return
52-week range
AS
ASML
LONG
ASML Holding N.V.
+25.7%
EV-weighted return
52-week range
ME
META
WATCH
Meta Platforms, Inc.
+10.7%
EV-weighted return
52-week range
KL
KLAC
LONG
KLA Corporation
+24.1%
EV-weighted return
52-week range
MS
MSFT
LONG
Microsoft Corporation
+24.8%
EV-weighted return
52-week range
IN
INTC
LONG
Intel Corporation
+36.8%
EV-weighted return
52-week range
Daily memo
Apr 15, 2026 · 00:21 EDT
22
Bullish
0
Bearish
19
Neutral
41
Tickers
360
Changes
Top scenario movers
TSM
+12pp
Bull → 83%
LULU
+10pp
→ 33%
HOOD
+8pp
→ 43%
JNJ
+8pp
→ 50%
GE
+8pp
→ 82%
MSFT
+8pp
→ 48%
PANW
+8pp
→ 60%
TSLA
+6pp
Bull → 28%
Net direction
Bullish22
Neutral19
Bearish0
By action type
Scenario probability163
Flag for review46
No action151
Knowledge map
54 companies · 325 edges
ForceOrbit
NVDA AVGO AMD MRVL KLAC ARM AMAT MU GOOGL NFLX JNJ MRK XOM BAC MA COST CEG
Info Tech Comms Financials Healthcare Energy Cons Def Utilities
Source hierarchy
Every claim carries its tier
The system distinguishes between what's verified from primary filings and what's inferred by an LLM — and surfaces that distinction in every output.
01
Primary data
Verbatim filings and authoritative economic releases. Zero interpretation.
Regulatory filings Fundamentals Macro indicators Market pricing
98%
Confidence
02
Derived analytics
Ratios, sentiment scores, and technicals computed from primary inputs.
NFSI sentiment Cluster momentum EV weightings Technical setups
90%
Confidence
03
Narrative signals
News flow, transcript tone, and management language — scored and decay-weighted.
News routing Transcripts Tone scoring
68%
Confidence
04
LLM inference
Synthesis, thesis writing, and scenario reasoning — always grounded in the tiers above.
Thesis synthesis Scenario narration Pivot framing
55%
Confidence
End-to-end pipeline
How data becomes conviction
Every stage writes back into context, so outputs compound rather than restart from zero.
4 stages
Continuous
Traceable
01
Data
Filings · Fundamentals
Transcripts
News · Macro
6 source types
02
ace · context engine
Company records
Knowledge graph
Source hierarchy
49 entities · 325 edges
03
Agent systems
News · Earnings
Sentiment · Monitor
Decision · Temporal
6 agents · parallel
04
Output
Thesis · Scorecard
Earnings briefs
Daily memo
5 deliverables
↺
Outputs feed back into context — the system learns from its own research
Live now on Autopilot

Follow our system-driven portfolio.

A live portfolio built directly from the research system's insights. Every position and every rebalance is driven by the agents — not by analyst discretion.

Check out our Portfolio on Autopilot

Seen enough? Get in early.

Alpha Foundation is opening access in stages. Drop your email and we'll reach out when your seat is ready.

Sign up for updates → Contact us →
Tweaks αf
Adjust visual defaults. Changes persist.