Weekly nutrition & training coach driven by InBody 770 scans

fis mcp 17 June 2026

COMPETITIVE LANDSCAPE (verified, June 2026) The space splits into 3 layers; almost every product sits in only 1–2: 1) Hardware scanners (data source): DEXA (gold standard, X-ray, clinic, ~$40–150); InBody (BIA, gym/clinic); Visbody/Fit3D (3D-optical + BIA kiosks, $5–15k); Withings Body Scan (premium scale). 2) Phone-scan apps: ZOZOFIT (3D scan + AI food scanner + meal plans), FitCommit ($3.99/mo body-fat + macros), Spren (peer-reviewed, sold as B2B API), MeThreeSixty, Body Snap. NOTE: camera-based — physically cannot see water balance or phase angle. 3) AI nutrition coaches: Welling.ai, Nutrizor, Strongr Fastr (auto-adjusts weekly), Fitbod. They start from self-reported weight, NOT a body-composition scan. Closest to the full loop: - Hume Health Body Pod — right scan (8-electrode multi-freq BIA) but soft, unstructured advice. - Zing Coach — weekly adapting program but camera scan (not BIA) + weak nutrition. - InBody App — owns the scan + data but generates no program. - Strongr Fastr — weekly, both sides, but blind to composition (optimizes scale weight; can't tell muscle loss from fat loss). Conclusion: nobody owns scan + weekly program + BOTH nutrition & training + composition-aware. That intersection is empty.

fis mcp 17 June 2026

INBODY 770 — INPUT DEVICE SPEC (verified) Method: DSM-BIA, 8-point tactile electrodes (barefoot standing + hand electrodes). Multi-frequency: impedance at 6 frequencies (1, 5, 50, 250, 500, 1000 kHz) across 5 segments (R/L arm, trunk, R/L leg) = 30 impedance measurements. Reactance at 3 freq (5/50/250 kHz) → phase angle. No empirical estimation (no age/sex/ethnicity back-fill). ~30–60s test. Two result sheets: body composition + body water. Full output = the app's input schema: - Core: TBW split into ICW + ECW; protein, minerals, dry lean mass, body fat mass; SMM, soft lean mass, weight, BMI, PBF. - Signature signals camera/scale apps CANNOT get: ECW/TBW ratio (whole + per-segment; healthy 0.360–0.390); whole-body + segmental phase angle; reactance; segmental lean analysis; segmental ICW/ECW. - Derived: visceral fat area & level, BMR, WHR, body cell mass, InBody score, target weight/muscle/fat. - Raw: the 30 segmental impedance values, available digitally. Sources: inbodyusa.com/general/770-result-sheet-interpretation, inbodyusa.com/general/technology

fis mcp 17 June 2026

DATA INTEGRATION PATH — LookinBody Web API The 770 syncs tests to LookinBody Web (cloud). Third-party apps pull via REST: - Auth: two headers — Account (LookinBody Web credentials) + API-KEY (generated in account settings). Missing key → 401. - Scope: UserID = single location; UserToken = data across ALL connected locations (use for a multi-gym product). - Endpoint: https://{webapiaddress}/user/test returns test records. Webhooks fire on new tests → app can auto-generate the new week the moment someone scans. - Region endpoints: apiusa.lookinbody.com (US), apieur.lookinbody.com (EU). Paid subscription; submit the LookinBody API Request form to get a key. Other tiers: LB120 (local PC DB), LB Integration (direct system/EMR). Result sheet (PDF/QR) = manual; ignore for automation. Scan→program trigger flow: new 770 scan → LookinBody Web → webhook → weekly engine → nutrition + training week. No hardware to build; the device + API already exist. Sources: apiusa.lookinbody.com, inbodyusa.com/web-api, lbwebfaq.inbodyusa.com (3rd-party integration guide)

fis mcp 17 June 2026

MARKET SIZING (2024–2030) Body composition analyzers: ~$1.4–2.0B in 2024; ~8.5% CAGR; reaching ~$2.05B by 2030 (some firms cite ~$619M for a narrower segment). Driver: ~1B people projected to live with obesity by 2030. Diet & nutrition apps: ~$2.1–5.8B (2024/25, varies by firm); ~12–13.4% CAGR; reaching ~$4.5–10B by 2030. North America ~36% revenue share. Demand signal (qualitative but strong): multiple gyms publish "paste your InBody numbers into ChatGPT" guides — users are manually duct-taping the scan→plan workflow because no product does it end-to-end. That manual workaround is the clearest evidence of unmet demand. Sources: Strategic Market Research, Grand View Research, Allied Market Research, Mordor Intelligence.

fis mcp 17 June 2026

SCAN ACCURACY CONTEXT (why BIA/770 over camera scans) Smartphone 3D scans are now validated vs DXA (Tinsley et al. 2024, n=131): statistically equivalent within ±2%, ICC 0.996–0.997, ~3.5–4.2% mean absolute error — good enough for trend tracking, not clinical diagnosis. BUT camera scans only capture shape; they cannot measure intra/extracellular water or phase angle. Why this matters for the moat: the 770's multi-frequency current penetrates cell membranes differently at each frequency, separating ICW vs ECW. That yields recovery/hydration/cellular-integrity signals (phase angle, ECW/TBW) which drive BOTH the meal and the training prescription. Camera-based competitors (Zing, FitCommit, ZOZOFIT) physically cannot replicate this — it is the structural data advantage. Caveat: BIA is hydration-sensitive, so standardize scan conditions (time of day, hydration, pre-/post-meal) and frame outputs as trend-based, not absolute. Source: PMC11491362 (smartphone 3D imaging vs DXA validation).

fis mcp 17 June 2026

Section: design Proposal: WEEKLY ADJUSTMENT ENGINE (the moat) Input each week: fresh 770 scan {fat mass, lean mass, BF%, ICW/ECW, ECW/TBW, phase angle, segmental lean, RMR} + logged adherence from the prior week. Logic: 1) Read the DELTA, not the absolute. Lost weight but lean mass dropped + ECW fell → deficit too aggressive / under-recovered → ease deficit, lower training volume. A weight-only app (e.g. Strongr Fastr) would celebrate that same week. 2) Prescribe the week, both sides: - Nutrition: weekly kcal/macro target → 7-day meal plan + grocery list; protein floored to lean mass (~2g/kg) to protect muscle. - Training: split/volume/intensity modulated by phase angle (recovery readiness); bias toward lagging segments from the segmental scan. 3) Close the loop on re-scan: next 770 measures whether lean mass actually held; model learns this individual's response rate and recalibrates → personalized, not generic. Split of responsibility: the engine = a real adaptive energy-balance + lean-preservation model that computes the numbers. The LLM = writes the meal plans and the plain-language weekly explanation, and never invents figures. Rationale: Captures the differentiating mechanism (composition-aware weekly adaptation using 770 signals) as a concrete, buildable spec rather than a tagline.

fis mcp 17 June 2026

Section: validation Proposal: VALIDATION PLAN 1) Confirm LookinBody Web API access + pricing (Account + API-KEY headers, /user/test endpoint, new-scan webhook). Build a read-only ingest against one real 770 feed. 2) Spec the weekly engine decision table: scan-delta thresholds → calorie/macro/training-volume changes. Back-test it against a few coached clients' historical 770 scan series. 3) Pilot with ONE gym/clinic running a 770: members connect scans, receive a weekly plan; measure adherence and 6–12 week composition change vs a control group. 4) Wedge test: a "Connect your InBody scan, get your week" landing page → measure conversion from existing 770-scanning members. Note: real composition change typically only shows at 6–12 weeks, so design the pilot around that horizon and use weekly scans for in-between adaptation rather than outcome judgement. Rationale: Turns the next-step line into a sequenced, falsifiable validation path with a concrete first integration and a real-world pilot.

fis mcp 17 June 2026

Section: risk Proposal: RISKS / GUARDRAILS - Incumbent: InBody's own Diet Guide (a one-shot BMR macro calculator) — the product must clearly exceed it via real meal plans + weekly adaptation + use of the 770's signature signals (ECW/TBW, phase angle, segmental lean). If it only outputs macros, there is no wedge. - Regulatory: keep wellness/educational framing, prominent "not medical advice"; use conservative deficits; flag underweight / eating-disorder patterns. - Accuracy ceiling: BIA is hydration-sensitive — standardize scan conditions and present outputs as trends, not absolutes. - Dependency / channel risk: relies on paid LookinBody Web API access and gyms/clinics as the scan channel — partner concentration risk. - Competitive: well-funded camera-scan players (Zing, FitCommit, ZOZOFIT) are moving toward the same scan→weekly-plan loop and could approximate it without the BIA depth. Rationale: Consolidates the verified competitive, regulatory, technical, and dependency risks surfaced during research into one guardrail set.

fis mcp 17 June 2026

Section: prototype Proposal: SMALLEST USEFUL PROTOTYPE Scope: a thin vertical slice — ONE goal (fat loss) and ONE re-scan cycle. 1) Ingest: read-only pull of a single 770 record from the LookinBody Web API. To de-risk API approval timing, the pilot can also accept a manually uploaded result-sheet PDF/CSV. 2) Engine v0: a deterministic decision table mapping {fat-mass delta, lean-mass delta, ECW/TBW shift, phase angle} → one of ~6 weekly states (hold / ease deficit / deepen deficit / add recovery / increase volume / deload). 3) Output: an LLM renders the chosen state into a 7-day meal plan + grocery list + a training week, plus a plain-language "why this week" note tied to the scan deltas. Demo path: connect/paste a real 770 scan → full week generated in <60s → upload next week's scan → plan visibly changes. Test data: 3–5 anonymized real 770 scan series (2+ scans each) from a partner gym to tune thresholds. No app store, no accounts — a single web page proves the loop. Rationale: Fills the empty Prototype chapter with a concrete smallest-slice build, demo path, and test-data requirement.