Automated Research Report India: 18 Nodes to a 20-50 Page Narrative

An automated research report India is not a prettier export of your questionnaire; it is a narrative organised around why you asked. On Hercules Works, that 'why' is the Research Brief — free-text business goals you write when creating the survey, stored in survey_chats.research_brief and injected into every major LLM prompt in the pipeline. The brief touches five places: Report Planner decides which questions matter, Insight Generator frames every finding against goals, Narrative Synthesis writes one goal narrative per objective, Report Drafter uses it as 'Research Context' for every section, and Section Auditor checks whether sections actually answer it. That is why a report on a pricing study does not start with demographics; it starts with the decision you face — 'should we launch at ₹499 or ₹699 for Tier 2 mothers?' — and then recruits evidence. Built by Jupiter Meta Labs in Bangalore, this North Star design is why automated research report India on Poseidon feels consultative, not clerical.

The report also draws on five context sources, not just response data. Source 1 is the Research Brief (the why), Source 2 is Survey Version/Questionnaire from survey_chats.survey_version (what was asked — text, types, options, scales), Source 3 is Survey Response Data at data/{chat_id}.parquet (what people said — every answer, DuckDB-optimised columnar), Source 4 is Audience Payload from survey_chats.audience_payload (who was targeted — demographics, quotas, geography via SuperJ), and Source 5 is Survey Intelligence computed at runtime (auto-discovered routing pairs via DuckDB null-correlation and goal to question mapping). A senior analyst in Delhi juggles these five mentally; Poseidon does it structurally, so recommendations are grounded in who was asked, what was asked, and what was found. That triangulation is how automated research report India avoids generic insights and delivers implications a CEO can sign.

Writing follows the 'So What' rule: every paragraph answers 'so what does this mean for the business?' as business implication first, then evidence with percent and N, never the reverse. BAD: '65% prefer digital platforms.' GOOD: 'Digital platforms are the dominant channel — 65% prefer them (n=612), making mobile-first features the highest-ROI investment for customer acquisition.' The drafting LLM is instructed to use structured HTML classes and to embed chart tokens like {{CHART_task_1_overall}} where a chart explains an insight, not as decoration. For a tiered India, that matters: a Tier 2 versus metros cut is not reported as two numbers; it is reported as an implication — 'Tier 2 is more price-elastic, so wallet cashback, not status, should lead creative in Lucknow' — with the chart as proof. That narrative discipline turns automated research report India on Hercules from data dump into decision memo, in hours not weeks.

Intelligence and Planning — three nodes — decide what to say and in what order. Survey Intelligence auto-detects routing pairs via DuckDB null-correlation, so conditional findings get the correct base (e.g., 'Among theatre-goers (n=40), 62% prefer butter popcorn') and each research goal is mapped to specific questions via LLM, ensuring coverage. Report Planner parses the questionnaire, identifies question types, demographic columns and cross-question linkages, and generates the ordered task list, deciding which questions are key metrics and which belong in the appendix. Analytics Planner then adds high-leverage derived tasks on top — weighted attribute scores, funnels, driver analysis, importance times performance matrices — so the report includes senior-analyst techniques, not just frequencies. Together, these three nodes ensure automated research report India on Poseidon is scoped by your Research Brief and questionnaire logic, not by a template that treats every survey as 40 equal charts.

Data and Visualisation — three nodes — do the math and choose the picture. Data Analysis executes DuckDB SQL per task — frequencies, means, cross-tabs, demographic cuts — respecting the Survey Knowledge Graph's column to formula mapping so multi-select uses UNNEST(string_split(col, ';')) and ranking uses Borda count. Open-Ended Intelligence codes free-text verbatims across 8 plus Indian languages (with Hinglish) when present, otherwise it is a no-op; theme extraction and sentiment are then available as nodes for synthesis. Visualisation generates Plotly charts (HTML plus PNG) with a chart-advisor LLM that picks the optimal type from data shape: horizontal bar for NPS and multi-select, grouped bar for grid/battery, ranking chart for ranking, donut for binary, pie for three categories, bar or horizontal for four to eight, horizontal bar for nine plus, heatmap/stacked bar for cross-tab, histogram for continuous, horizontal funnel for funnel, and horizontal bar with average rank for Borda/MaxDiff. All use a colourblind-aware Okabe-Ito inspired palette, Inter font and direct labelling. That is automated research report India where charts are authored, not auto-pasted.

In practice, this means Sneha's Hyderabad pricing study gets Van Westendorp curves as horizontal line charts with direct labels, NPS by city tier as a heatmap, and driver importance as a horizontal bar ranked by Shapley value — each chart tied to a task and referenced by a placeholder token that is only rendered if the data supports it. No chart for chart's sake, no 40-slide dump. Because everything is DuckDB-on-Parquet at 10-20× faster than Pandas and Parquet is columnar, even a 50,000-response survey's 40-column task list executes without loading the entire file, keeping report generation quick and cheap. Combined with the Research Brief as North Star and five context sources, the first six nodes ensure automated research report India starts with correct scope, correct math, and correct visuals before a single narrative sentence is written — the foundation a senior analyst would demand before drafting.

Intelligence and Narrative — four nodes — turn numbers into story, still anchored to your Research Brief. Insight Generation extracts per-question and cross-question findings into structured insight packs with severity ratings, so 'Tier 2 price sensitivity' is tagged as high severity for a Tier 2 launch decision. Narrative Synthesis writes three connected outputs: routing-conditional insights (e.g., 'among churners'), goal narratives (one per research objective, synthesising tasks that advance that goal), and demographic threads (consistent segment patterns across goals, like 'Tier 2 mothers are consistently more value-sensitive'). Goal Traceability tags every section with the goal(s) it advances, producing a coverage map that the auditor later checks — did every objective get evidence? Cross-Section Synthesis spots cross-cutting patterns spanning two plus sections, such as 'durability concerns in open text correlate with NPS detractors in Tier 2,' which no single frequency could surface. This is the core of automated research report India on Poseidon: narrative that is cross-referenced, not chapterised, so the CEO sees patterns, not just pages.

The payoff is a report structure that mirrors a consulting deliverable, not a survey tool export: Executive Summary, Research Goals and Decision Context, Methodology, Data Quality and Respondent Validation, Respondent Profile, Goal-wise Analysis, Key Question Findings, Cohort/Segment Insights, Strategic Synthesis, Recommendations, Conclusion and Appendix — with charts embedded only where they explain an insight. For a Mumbai fintech's UPI pricing study, that means the Van Westendorp Price Sensitivity Meter story is told once as a goal narrative ('where should we price to maximise adoption in Tier 1 versus Tier 2?'), not scattered as four separate percentages. For a Bangalore D2C concept test, 'purchase intent by concept and city tier' is synthesised with open-ended themes ('natural ingredients praised in Hindi verbatims') into one recommendation: 'Lead with concept B in Tier 2, but reframe durability for Tier 1.' That is automated research report India that writes around decisions, respecting the Research Brief as North Star at every prompt.

To keep the story honest, every insight pack carries its base N, denominator correction for skip logic, and confidence medium/high, so conditional insights correctly say 'n equals 40, not 400.' The graph's routing edges ensure the narrative never divides by the wrong base, and the five context sources ensure audience and questionnaire context travels with the numbers. A Government of Karnataka citizen survey can thus report 'water supply complaints in north Bangalore precede city-wide satisfaction drops by one quarter' with temporal synthesis from cross-section analysis, not just a bar chart. Built by Jupiter Meta Labs in Bangalore, this phase is why automated research report India on Hercules Works reads like a senior analyst spent a week connecting dots — except Poseidon did it in minutes, via LangGraph and Gemini, verified per number before drafting begins.

Drafting and Quality — four nodes — write and harden the report so it is client-ready. Report Drafting writes each section chunk-by-chunk via LLM, embedding chart tokens ({{CHART_task_1_overall}}) and using structured HTML classes for layout, so a chart placeholder resolves to Plotly HTML/PNG only if the data task succeeded and the narrative references it. Self-Consistency checks that numeric claims in synthesis sections match per-section data — if Strategic Synthesis says '62%' but Key Findings says '60.1%,' it flags mismatch. Section Audit runs a two-part audit: deterministic claim verification (re-compute numbers via fresh DuckDB SQL, as in chat) plus LLM writing-quality audit for jargon, hallucinations and business framing against the Research Brief. Safety Check validates PII leaks, numeric mismatches, coverage gaps and format errors. Any section scoring NEEDS_REVISION is sent back to drafting with specific instructions, up to two retry cycles per loop. Two conditional retry loops — Section Audit to Redraft and Safety Check to Redraft — ensure automated research report India on Poseidon is not first-draft thinking but audited thinking.

This quality stack is where Hercules Works diverges from generic 'AI report' tools that run one prompt and pray. On Poseidon, critical failure at any node terminates the pipeline with error logs surfaced in Prometheus metrics and /health/stats, so you know why a report did not generate rather than receiving a hallucinated PDF. Non-critical safety issues trigger re-draft and re-check, exactly as a Delhi agency's senior reviewer would mark up a junior's deck with red pen — 'add base N,' 'remove causal claim,' 'rephrase paisa vasool implication.' Because DuckDB reads Parquet columnarly and semantic cache at 78% hit rate avoids re-computing task results already verified in chat, drafting is fast even for 40-task reports. The result is automated research report India that is not just assembled but assured, with each paragraph earning its evidence before it is allowed to stay.

For Sneha in Hyderabad, that means when the 32-page report lands, the Pricing Research section's Van Westendorp numbers have already been re-derived, the NPS by city tier heatmap sums correctly, and the Recommendations cite the goal coverage map so nothing promised in the brief is missing. Charts are colourblind-safe, fonts are Inter, backgrounds transparent, and labelling direct — print-ready via WeasyPrint without a designer. Pricing at Free ₹0/month for first experiments through Pro ₹30,000/quarter (₹24,000 annual) for weekly reports makes this quality accessible without agency retainer. And like chat, report generation streams progress and persists: if you navigate away, the 18-node job continues and you return to a ready PDF. That is automated research report India built for Bharat's pace and scrutiny — Ekdum solid, from Bangalore, for decisions that cannot wait 6-8 weeks.

Assembly — two nodes — deliver the artefact you can send. Report Validation does a structural completeness check: are all required sections present — cover page, methodology, goal narratives, recommendations, appendix? Are chart tokens resolved and do referenced charts exist? Report Finalisation assembles the cover page, resolves chart placeholders into Plotly HTML/PNG, renders Markdown to HTML to PDF, and persists the three formats together with the chat turn in Postgres (Hercules shares Postgres via ADD COLUMN IF NOT EXISTS safe migrations). You get Markdown as canonical source, interactive HTML for browser review with embedded Plotly, and print-ready PDF for the board. Because everything is versioned with the chat, a re-run after new responses shows a diff, not a duplicate. That is automated research report India that ends with a deliverable, not a dashboard you must screenshot into slides.

Consider the workflow end-to-end for a real India brand. Responses flow from SuperJ (20M plus verified Indians, 60-90%+ rates, WhatsApp-native, 8 plus languages) as JSON, are typed, normalised and stored as Parquet, understood via the Survey Knowledge Graph, and then every question you ask in English traverses the same graph before any report task runs. The 18-node generator plans from your Research Brief, executes DuckDB SQL per task, codes open-ends, picks chart types via chart advisor, synthesises goal narratives around decisions, audits via three-layer verification and Self-Critique, and finalises Markdown/HTML/PDF. From JSON ingestion through CSV to Parquet to PDF, no manual Excel, no emailing sheets named FINAL_final_v3, no waiting for a designer. Hours, not weeks; 10-100× cheaper than legacy players who charge ₹10-50 lakhs per deck.

That is why Unilever, Kantar, Government of Karnataka, ICICI Prudential and SBI Mutual Fund trust Hercules Works for automated research report India: every claim has a verification trail, every chart has a token, every section traces to a goal, and every PDF is client-presentable without rework. Whether you test a new pack in Indore, a price in Jaipur, or a concept in Kochi, you get a 20-50 page narrative that connects data to decision with the 'So What' rule, authored around your Brief, and delivered at Free ₹0/month to start, Starter ₹1,119/month (₹895 annual, 20% off) and Pro ₹30,000/quarter (₹24,000 annual, 20% off). Start at hercules.works/ai, write a Brief that states the bet, and watch Poseidon turn SuperJ truth into a report your leadership will actually read and act on — in hours, not weeks, and paisa vasool.