Fixed Parallel Branches

Overview

Why this option ships first and what it achieves.

Option 2 implements the VC Deep Research Agent using LangGraph's fixed addEdge() branches. All 8 specialist agents are wired at graph-build time. Optional agents always execute — they return status: "no_data" early when public data is unavailable rather than being skipped.

The pipeline accepts a pitch deck (PDF/PPT), company name, or URL and produces a structured hybrid report for VC analysts in under 20 minutes. It maps to Stages 02–04 of the One Agentic six-stage interaction loop: Deal Intake → Enrichment + Thesis Evaluation → Conflict Surfacing.

Design Principles

• Transparency first: every data gap, conflict, and limitation is surfaced explicitly — never silently omitted
• Append-only state: each pipeline stage adds fields, never overwrites prior data — full audit trail per run
• Graceful degradation: optional agents and rate-limited tools degrade to partial results, not failures
• Zero-cost failures: credits deducted only on successful report generation
• Consistency: the same company run twice should produce materially equivalent findings

System Architecture

Seven layers from raw input to streamed report.

Layer Overview

Graph Topology

Graph Wiring

// graph.ts — the ~20 lines that differ from Option 1
const firstWave: AgentName[] = [
  "company", "founders", "market", "digital", "news", "legal", "financial"
];

// Orchestrator fans out to all 7 non-claims agents simultaneously
firstWave.forEach(a => graph.addEdge("orchestrator", a));

// All 7 converge at barrier before Claims runs
firstWave.forEach(a => graph.addEdge(a, "barrier"));

// Sequential tail: Claims → Reducer → Report
graph.addEdge("barrier", "claims");
graph.addEdge("claims", "reducer");
graph.addEdge("reducer", "report");

State Schema

Append-only TypeScript interfaces that flow through the entire graph.

ResearchState

interface ResearchState {
  // ── Input ──────────────────────────────────────────────────────────────
  runId:           string;
  fundThesis:      FundThesis;        // from Stage 01 thesis capture
  rawInput:        RawInput;          // uploaded file, company name, or URL

  // ── Extracted by Orchestrator ─────────────────────────────────────────
  companyProfile:  CompanyProfile;    // normalised entity
  claimedMetrics:  ClaimedMetric[];   // what the deck says
  enabledAgents:   AgentName[];       // all 8 in Option 2

  // ── Agent outputs (Annotated reducers — parallel-safe) ────────────────
  companyFindings:   AgentResult;
  founderFindings:   AgentResult;
  marketFindings:    AgentResult | null;   // optional
  claimsFindings:    AgentResult;
  digitalFindings:   AgentResult;
  newsFindings:      AgentResult;
  legalFindings:     AgentResult | null;   // optional
  financialFindings: AgentResult | null;   // optional

  // ── Reducer outputs ───────────────────────────────────────────────────
  conflicts:       Conflict[];
  thesisFitScore:  number;            // 0–100
  report:          HybridReport | null;
}

Supporting Types

// Every agent returns this — agents never throw
interface AgentResult {
  status:     "ok" | "partial" | "no_data";
  confidence: number;        // 0–1
  summary:    string;        // prose for the report section
  dataPoints: DataPoint[];
  sources:    Source[];
  flags:      Flag[];        // agent-local red flags
  durationMs: number;
}

// Conflict surfaced by Reducer
interface Conflict {
  type:           "implausible_absence" | "manufactured_signal" | "cross_source_conflict";
  severity:       "high" | "medium" | "low";
  description:    string;
  sourceA:        string;
  sourceB:        string;
  agentsInvolved: AgentName[];
}

// Normalised company entity
interface CompanyProfile {
  name:         string;
  domain?:      string;
  founderNames: string[];
  sector?:      string;
  stage?:       string;       // "pre-seed" | "seed" | "series-a"
  hq?:          string;
  linkedinUrl?: string;
  githubOrg?:   string;
}

// A claim extracted from the pitch deck
interface ClaimedMetric {
  field:          string;     // e.g. "ARR", "MAU"
  value:          string;     // e.g. "$2.4M"
  sourceSlide:    number | null;
  validated?:     boolean;    // filled by Claims agent
  contradiction?: string;
}

// All specialist agents conform to this signature
type SpecialistAgent = (state: ResearchState) => Promise<Partial<ResearchState>>;

State Flow

Input Parsing

Four input types normalise to the same CompanyProfile + ClaimedMetric[] before agents run.

Extraction Prompt Contract

const EXTRACTION_SYSTEM_PROMPT = `
You are extracting structured data from a startup document.
Return ONLY valid JSON matching the CompanyProfile + ClaimedMetric[] schema.
For each metric, record: field name, stated value, and slide/page number if visible.
If a field is not present in the document, omit it — do not infer or hallucinate.
Mark any metric you are uncertain about with "confidence": "low".
`;
// Enforced via Claude tool_use / structured output — no free-text response

Ambiguous Input Handling

Agent Tool Registry

Eight specialist agents, their tool access, and tier costs.

All tools are LangGraph ToolNodes wrapping typed async functions. Every tool uses fetchWithRetry — exponential backoff, 3 attempts, max 30s. Tools never throw; errors return empty DataPoint[].

T0 Free T1 Metered T2 Subscription

Required Agents

Optional Agents (return no_data gracefully)

Barrier Node Pattern

How Claims gets complete state from all 7 first-wave agents.

The Claims agent must receive complete outputs from all 7 first-wave agents before it can detect cross-source conflicts and implausible absences. LangGraph does not natively support "wait for N concurrent nodes" — this is resolved with a synchronisation checkpoint node.

// barrier.ts
function barrierNode(state: ResearchState): Partial<ResearchState> {
  // LangGraph's Annotated reducer holds this node until all 7 incoming
  // edges have written to state. Timed-out agents fill their field with
  // a "partial" stub before the barrier receives control.
  return {}; // synchronisation point only — no state mutation
}

Timeout handling: required agents that exceed 3 minutes return a status: "partial" stub result. The barrier never waits indefinitely — it proceeds once all branches have written (stub or real).

Hybrid Report & SSE Streaming

Four report zones populated progressively as agents complete.

Report Zone Structure

SSE Event Schema

Zone 1 renders within approximately 5 seconds of run start. The VC sees live progress rather than a waiting spinner.

Azure Deployment

Six Azure services and how they connect.

Request Flow

1. VC uploads file or enters company name → Blob Storage + Service Bus enqueues run
2. Container Apps picks up message → Document Intelligence extracts text → Orchestrator initialises ResearchState
3. 7 first-wave agents run in parallel → SSE stream sends live progress events to browser
4. Barrier waits for all 7 → Claims agent runs → Reducer merges results + detects conflicts
5. HybridReport generated → persisted to Cosmos DB → HTML served from Static Web Apps
6. Full report available within target of < 20 minutes from upload

Error Handling & Graceful Degradation

Every failure mode and what the VC sees.

Confidence Score Rules

Implementation Plans

Three sequential plans covering the full build. Each is independently testable.

Open Questions & Migration

Resolved questions and the path to Option 1.

Migration to Option 1

When all 8 agents are validated on real deal data, migrate to Option 1 (dynamic Send API) by replacing graph wiring only. No agent files change.

• What changes: ~20 lines in graph.ts + ~10 lines in orchestrator.ts
• What does NOT change: all 8 agents, all 12 tools, state schema, reducer, SSE server, Azure deployment, report frontend
• Rollback: restore the addEdge() block in graph.ts — no other files need reverting

See option1.html for the complete Option 1 specification and migration checklist.